xref: /openbmc/linux/arch/powerpc/kvm/book3s_hv.c (revision e4781421e883340b796da5a724bda7226817990b)
1 /*
2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Paul Mackerras <paulus@au1.ibm.com>
7  *    Alexander Graf <agraf@suse.de>
8  *    Kevin Wolf <mail@kevin-wolf.de>
9  *
10  * Description: KVM functions specific to running on Book 3S
11  * processors in hypervisor mode (specifically POWER7 and later).
12  *
13  * This file is derived from arch/powerpc/kvm/book3s.c,
14  * by Alexander Graf <agraf@suse.de>.
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License, version 2, as
18  * published by the Free Software Foundation.
19  */
20 
21 #include <linux/kvm_host.h>
22 #include <linux/err.h>
23 #include <linux/slab.h>
24 #include <linux/preempt.h>
25 #include <linux/sched.h>
26 #include <linux/delay.h>
27 #include <linux/export.h>
28 #include <linux/fs.h>
29 #include <linux/anon_inodes.h>
30 #include <linux/cpu.h>
31 #include <linux/cpumask.h>
32 #include <linux/spinlock.h>
33 #include <linux/page-flags.h>
34 #include <linux/srcu.h>
35 #include <linux/miscdevice.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/reg.h>
39 #include <asm/cputable.h>
40 #include <asm/cacheflush.h>
41 #include <asm/tlbflush.h>
42 #include <linux/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/kvm_ppc.h>
45 #include <asm/kvm_book3s.h>
46 #include <asm/mmu_context.h>
47 #include <asm/lppaca.h>
48 #include <asm/processor.h>
49 #include <asm/cputhreads.h>
50 #include <asm/page.h>
51 #include <asm/hvcall.h>
52 #include <asm/switch_to.h>
53 #include <asm/smp.h>
54 #include <asm/dbell.h>
55 #include <asm/hmi.h>
56 #include <asm/pnv-pci.h>
57 #include <asm/mmu.h>
58 #include <asm/opal.h>
59 #include <asm/xics.h>
60 #include <linux/gfp.h>
61 #include <linux/vmalloc.h>
62 #include <linux/highmem.h>
63 #include <linux/hugetlb.h>
64 #include <linux/kvm_irqfd.h>
65 #include <linux/irqbypass.h>
66 #include <linux/module.h>
67 #include <linux/compiler.h>
68 #include <linux/of.h>
69 
70 #include "book3s.h"
71 
72 #define CREATE_TRACE_POINTS
73 #include "trace_hv.h"
74 
75 /* #define EXIT_DEBUG */
76 /* #define EXIT_DEBUG_SIMPLE */
77 /* #define EXIT_DEBUG_INT */
78 
79 /* Used to indicate that a guest page fault needs to be handled */
80 #define RESUME_PAGE_FAULT	(RESUME_GUEST | RESUME_FLAG_ARCH1)
81 /* Used to indicate that a guest passthrough interrupt needs to be handled */
82 #define RESUME_PASSTHROUGH	(RESUME_GUEST | RESUME_FLAG_ARCH2)
83 
84 /* Used as a "null" value for timebase values */
85 #define TB_NIL	(~(u64)0)
86 
87 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
88 
89 static int dynamic_mt_modes = 6;
90 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
91 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
92 static int target_smt_mode;
93 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
94 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
95 
96 #ifdef CONFIG_KVM_XICS
97 static struct kernel_param_ops module_param_ops = {
98 	.set = param_set_int,
99 	.get = param_get_int,
100 };
101 
102 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
103 							S_IRUGO | S_IWUSR);
104 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
105 
106 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
107 							S_IRUGO | S_IWUSR);
108 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
109 #endif
110 
111 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
112 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
113 
114 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
115 		int *ip)
116 {
117 	int i = *ip;
118 	struct kvm_vcpu *vcpu;
119 
120 	while (++i < MAX_SMT_THREADS) {
121 		vcpu = READ_ONCE(vc->runnable_threads[i]);
122 		if (vcpu) {
123 			*ip = i;
124 			return vcpu;
125 		}
126 	}
127 	return NULL;
128 }
129 
130 /* Used to traverse the list of runnable threads for a given vcore */
131 #define for_each_runnable_thread(i, vcpu, vc) \
132 	for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
133 
134 static bool kvmppc_ipi_thread(int cpu)
135 {
136 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
137 
138 	/* On POWER9 we can use msgsnd to IPI any cpu */
139 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
140 		msg |= get_hard_smp_processor_id(cpu);
141 		smp_mb();
142 		__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
143 		return true;
144 	}
145 
146 	/* On POWER8 for IPIs to threads in the same core, use msgsnd */
147 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
148 		preempt_disable();
149 		if (cpu_first_thread_sibling(cpu) ==
150 		    cpu_first_thread_sibling(smp_processor_id())) {
151 			msg |= cpu_thread_in_core(cpu);
152 			smp_mb();
153 			__asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
154 			preempt_enable();
155 			return true;
156 		}
157 		preempt_enable();
158 	}
159 
160 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
161 	if (cpu >= 0 && cpu < nr_cpu_ids) {
162 		if (paca[cpu].kvm_hstate.xics_phys) {
163 			xics_wake_cpu(cpu);
164 			return true;
165 		}
166 		opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
167 		return true;
168 	}
169 #endif
170 
171 	return false;
172 }
173 
174 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
175 {
176 	int cpu;
177 	struct swait_queue_head *wqp;
178 
179 	wqp = kvm_arch_vcpu_wq(vcpu);
180 	if (swait_active(wqp)) {
181 		swake_up(wqp);
182 		++vcpu->stat.halt_wakeup;
183 	}
184 
185 	if (kvmppc_ipi_thread(vcpu->arch.thread_cpu))
186 		return;
187 
188 	/* CPU points to the first thread of the core */
189 	cpu = vcpu->cpu;
190 	if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
191 		smp_send_reschedule(cpu);
192 }
193 
194 /*
195  * We use the vcpu_load/put functions to measure stolen time.
196  * Stolen time is counted as time when either the vcpu is able to
197  * run as part of a virtual core, but the task running the vcore
198  * is preempted or sleeping, or when the vcpu needs something done
199  * in the kernel by the task running the vcpu, but that task is
200  * preempted or sleeping.  Those two things have to be counted
201  * separately, since one of the vcpu tasks will take on the job
202  * of running the core, and the other vcpu tasks in the vcore will
203  * sleep waiting for it to do that, but that sleep shouldn't count
204  * as stolen time.
205  *
206  * Hence we accumulate stolen time when the vcpu can run as part of
207  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
208  * needs its task to do other things in the kernel (for example,
209  * service a page fault) in busy_stolen.  We don't accumulate
210  * stolen time for a vcore when it is inactive, or for a vcpu
211  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
212  * a misnomer; it means that the vcpu task is not executing in
213  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
214  * the kernel.  We don't have any way of dividing up that time
215  * between time that the vcpu is genuinely stopped, time that
216  * the task is actively working on behalf of the vcpu, and time
217  * that the task is preempted, so we don't count any of it as
218  * stolen.
219  *
220  * Updates to busy_stolen are protected by arch.tbacct_lock;
221  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
222  * lock.  The stolen times are measured in units of timebase ticks.
223  * (Note that the != TB_NIL checks below are purely defensive;
224  * they should never fail.)
225  */
226 
227 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
228 {
229 	unsigned long flags;
230 
231 	spin_lock_irqsave(&vc->stoltb_lock, flags);
232 	vc->preempt_tb = mftb();
233 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
234 }
235 
236 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
237 {
238 	unsigned long flags;
239 
240 	spin_lock_irqsave(&vc->stoltb_lock, flags);
241 	if (vc->preempt_tb != TB_NIL) {
242 		vc->stolen_tb += mftb() - vc->preempt_tb;
243 		vc->preempt_tb = TB_NIL;
244 	}
245 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
246 }
247 
248 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
249 {
250 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
251 	unsigned long flags;
252 
253 	/*
254 	 * We can test vc->runner without taking the vcore lock,
255 	 * because only this task ever sets vc->runner to this
256 	 * vcpu, and once it is set to this vcpu, only this task
257 	 * ever sets it to NULL.
258 	 */
259 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
260 		kvmppc_core_end_stolen(vc);
261 
262 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
263 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
264 	    vcpu->arch.busy_preempt != TB_NIL) {
265 		vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
266 		vcpu->arch.busy_preempt = TB_NIL;
267 	}
268 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
269 }
270 
271 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
272 {
273 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
274 	unsigned long flags;
275 
276 	if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
277 		kvmppc_core_start_stolen(vc);
278 
279 	spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
280 	if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
281 		vcpu->arch.busy_preempt = mftb();
282 	spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
283 }
284 
285 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
286 {
287 	/*
288 	 * Check for illegal transactional state bit combination
289 	 * and if we find it, force the TS field to a safe state.
290 	 */
291 	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
292 		msr &= ~MSR_TS_MASK;
293 	vcpu->arch.shregs.msr = msr;
294 	kvmppc_end_cede(vcpu);
295 }
296 
297 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
298 {
299 	vcpu->arch.pvr = pvr;
300 }
301 
302 /* Dummy value used in computing PCR value below */
303 #define PCR_ARCH_300	(PCR_ARCH_207 << 1)
304 
305 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
306 {
307 	unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
308 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
309 
310 	/* We can (emulate) our own architecture version and anything older */
311 	if (cpu_has_feature(CPU_FTR_ARCH_300))
312 		host_pcr_bit = PCR_ARCH_300;
313 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
314 		host_pcr_bit = PCR_ARCH_207;
315 	else if (cpu_has_feature(CPU_FTR_ARCH_206))
316 		host_pcr_bit = PCR_ARCH_206;
317 	else
318 		host_pcr_bit = PCR_ARCH_205;
319 
320 	/* Determine lowest PCR bit needed to run guest in given PVR level */
321 	guest_pcr_bit = host_pcr_bit;
322 	if (arch_compat) {
323 		switch (arch_compat) {
324 		case PVR_ARCH_205:
325 			guest_pcr_bit = PCR_ARCH_205;
326 			break;
327 		case PVR_ARCH_206:
328 		case PVR_ARCH_206p:
329 			guest_pcr_bit = PCR_ARCH_206;
330 			break;
331 		case PVR_ARCH_207:
332 			guest_pcr_bit = PCR_ARCH_207;
333 			break;
334 		case PVR_ARCH_300:
335 			guest_pcr_bit = PCR_ARCH_300;
336 			break;
337 		default:
338 			return -EINVAL;
339 		}
340 	}
341 
342 	/* Check requested PCR bits don't exceed our capabilities */
343 	if (guest_pcr_bit > host_pcr_bit)
344 		return -EINVAL;
345 
346 	spin_lock(&vc->lock);
347 	vc->arch_compat = arch_compat;
348 	/* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
349 	vc->pcr = host_pcr_bit - guest_pcr_bit;
350 	spin_unlock(&vc->lock);
351 
352 	return 0;
353 }
354 
355 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
356 {
357 	int r;
358 
359 	pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
360 	pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
361 	       vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
362 	for (r = 0; r < 16; ++r)
363 		pr_err("r%2d = %.16lx  r%d = %.16lx\n",
364 		       r, kvmppc_get_gpr(vcpu, r),
365 		       r+16, kvmppc_get_gpr(vcpu, r+16));
366 	pr_err("ctr = %.16lx  lr  = %.16lx\n",
367 	       vcpu->arch.ctr, vcpu->arch.lr);
368 	pr_err("srr0 = %.16llx srr1 = %.16llx\n",
369 	       vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
370 	pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
371 	       vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
372 	pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
373 	       vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
374 	pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
375 	       vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
376 	pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
377 	pr_err("fault dar = %.16lx dsisr = %.8x\n",
378 	       vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
379 	pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
380 	for (r = 0; r < vcpu->arch.slb_max; ++r)
381 		pr_err("  ESID = %.16llx VSID = %.16llx\n",
382 		       vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
383 	pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
384 	       vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
385 	       vcpu->arch.last_inst);
386 }
387 
388 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
389 {
390 	struct kvm_vcpu *ret;
391 
392 	mutex_lock(&kvm->lock);
393 	ret = kvm_get_vcpu_by_id(kvm, id);
394 	mutex_unlock(&kvm->lock);
395 	return ret;
396 }
397 
398 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
399 {
400 	vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
401 	vpa->yield_count = cpu_to_be32(1);
402 }
403 
404 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
405 		   unsigned long addr, unsigned long len)
406 {
407 	/* check address is cacheline aligned */
408 	if (addr & (L1_CACHE_BYTES - 1))
409 		return -EINVAL;
410 	spin_lock(&vcpu->arch.vpa_update_lock);
411 	if (v->next_gpa != addr || v->len != len) {
412 		v->next_gpa = addr;
413 		v->len = addr ? len : 0;
414 		v->update_pending = 1;
415 	}
416 	spin_unlock(&vcpu->arch.vpa_update_lock);
417 	return 0;
418 }
419 
420 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
421 struct reg_vpa {
422 	u32 dummy;
423 	union {
424 		__be16 hword;
425 		__be32 word;
426 	} length;
427 };
428 
429 static int vpa_is_registered(struct kvmppc_vpa *vpap)
430 {
431 	if (vpap->update_pending)
432 		return vpap->next_gpa != 0;
433 	return vpap->pinned_addr != NULL;
434 }
435 
436 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
437 				       unsigned long flags,
438 				       unsigned long vcpuid, unsigned long vpa)
439 {
440 	struct kvm *kvm = vcpu->kvm;
441 	unsigned long len, nb;
442 	void *va;
443 	struct kvm_vcpu *tvcpu;
444 	int err;
445 	int subfunc;
446 	struct kvmppc_vpa *vpap;
447 
448 	tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
449 	if (!tvcpu)
450 		return H_PARAMETER;
451 
452 	subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
453 	if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
454 	    subfunc == H_VPA_REG_SLB) {
455 		/* Registering new area - address must be cache-line aligned */
456 		if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
457 			return H_PARAMETER;
458 
459 		/* convert logical addr to kernel addr and read length */
460 		va = kvmppc_pin_guest_page(kvm, vpa, &nb);
461 		if (va == NULL)
462 			return H_PARAMETER;
463 		if (subfunc == H_VPA_REG_VPA)
464 			len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
465 		else
466 			len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
467 		kvmppc_unpin_guest_page(kvm, va, vpa, false);
468 
469 		/* Check length */
470 		if (len > nb || len < sizeof(struct reg_vpa))
471 			return H_PARAMETER;
472 	} else {
473 		vpa = 0;
474 		len = 0;
475 	}
476 
477 	err = H_PARAMETER;
478 	vpap = NULL;
479 	spin_lock(&tvcpu->arch.vpa_update_lock);
480 
481 	switch (subfunc) {
482 	case H_VPA_REG_VPA:		/* register VPA */
483 		if (len < sizeof(struct lppaca))
484 			break;
485 		vpap = &tvcpu->arch.vpa;
486 		err = 0;
487 		break;
488 
489 	case H_VPA_REG_DTL:		/* register DTL */
490 		if (len < sizeof(struct dtl_entry))
491 			break;
492 		len -= len % sizeof(struct dtl_entry);
493 
494 		/* Check that they have previously registered a VPA */
495 		err = H_RESOURCE;
496 		if (!vpa_is_registered(&tvcpu->arch.vpa))
497 			break;
498 
499 		vpap = &tvcpu->arch.dtl;
500 		err = 0;
501 		break;
502 
503 	case H_VPA_REG_SLB:		/* register SLB shadow buffer */
504 		/* Check that they have previously registered a VPA */
505 		err = H_RESOURCE;
506 		if (!vpa_is_registered(&tvcpu->arch.vpa))
507 			break;
508 
509 		vpap = &tvcpu->arch.slb_shadow;
510 		err = 0;
511 		break;
512 
513 	case H_VPA_DEREG_VPA:		/* deregister VPA */
514 		/* Check they don't still have a DTL or SLB buf registered */
515 		err = H_RESOURCE;
516 		if (vpa_is_registered(&tvcpu->arch.dtl) ||
517 		    vpa_is_registered(&tvcpu->arch.slb_shadow))
518 			break;
519 
520 		vpap = &tvcpu->arch.vpa;
521 		err = 0;
522 		break;
523 
524 	case H_VPA_DEREG_DTL:		/* deregister DTL */
525 		vpap = &tvcpu->arch.dtl;
526 		err = 0;
527 		break;
528 
529 	case H_VPA_DEREG_SLB:		/* deregister SLB shadow buffer */
530 		vpap = &tvcpu->arch.slb_shadow;
531 		err = 0;
532 		break;
533 	}
534 
535 	if (vpap) {
536 		vpap->next_gpa = vpa;
537 		vpap->len = len;
538 		vpap->update_pending = 1;
539 	}
540 
541 	spin_unlock(&tvcpu->arch.vpa_update_lock);
542 
543 	return err;
544 }
545 
546 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
547 {
548 	struct kvm *kvm = vcpu->kvm;
549 	void *va;
550 	unsigned long nb;
551 	unsigned long gpa;
552 
553 	/*
554 	 * We need to pin the page pointed to by vpap->next_gpa,
555 	 * but we can't call kvmppc_pin_guest_page under the lock
556 	 * as it does get_user_pages() and down_read().  So we
557 	 * have to drop the lock, pin the page, then get the lock
558 	 * again and check that a new area didn't get registered
559 	 * in the meantime.
560 	 */
561 	for (;;) {
562 		gpa = vpap->next_gpa;
563 		spin_unlock(&vcpu->arch.vpa_update_lock);
564 		va = NULL;
565 		nb = 0;
566 		if (gpa)
567 			va = kvmppc_pin_guest_page(kvm, gpa, &nb);
568 		spin_lock(&vcpu->arch.vpa_update_lock);
569 		if (gpa == vpap->next_gpa)
570 			break;
571 		/* sigh... unpin that one and try again */
572 		if (va)
573 			kvmppc_unpin_guest_page(kvm, va, gpa, false);
574 	}
575 
576 	vpap->update_pending = 0;
577 	if (va && nb < vpap->len) {
578 		/*
579 		 * If it's now too short, it must be that userspace
580 		 * has changed the mappings underlying guest memory,
581 		 * so unregister the region.
582 		 */
583 		kvmppc_unpin_guest_page(kvm, va, gpa, false);
584 		va = NULL;
585 	}
586 	if (vpap->pinned_addr)
587 		kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
588 					vpap->dirty);
589 	vpap->gpa = gpa;
590 	vpap->pinned_addr = va;
591 	vpap->dirty = false;
592 	if (va)
593 		vpap->pinned_end = va + vpap->len;
594 }
595 
596 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
597 {
598 	if (!(vcpu->arch.vpa.update_pending ||
599 	      vcpu->arch.slb_shadow.update_pending ||
600 	      vcpu->arch.dtl.update_pending))
601 		return;
602 
603 	spin_lock(&vcpu->arch.vpa_update_lock);
604 	if (vcpu->arch.vpa.update_pending) {
605 		kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
606 		if (vcpu->arch.vpa.pinned_addr)
607 			init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
608 	}
609 	if (vcpu->arch.dtl.update_pending) {
610 		kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
611 		vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
612 		vcpu->arch.dtl_index = 0;
613 	}
614 	if (vcpu->arch.slb_shadow.update_pending)
615 		kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
616 	spin_unlock(&vcpu->arch.vpa_update_lock);
617 }
618 
619 /*
620  * Return the accumulated stolen time for the vcore up until `now'.
621  * The caller should hold the vcore lock.
622  */
623 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
624 {
625 	u64 p;
626 	unsigned long flags;
627 
628 	spin_lock_irqsave(&vc->stoltb_lock, flags);
629 	p = vc->stolen_tb;
630 	if (vc->vcore_state != VCORE_INACTIVE &&
631 	    vc->preempt_tb != TB_NIL)
632 		p += now - vc->preempt_tb;
633 	spin_unlock_irqrestore(&vc->stoltb_lock, flags);
634 	return p;
635 }
636 
637 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
638 				    struct kvmppc_vcore *vc)
639 {
640 	struct dtl_entry *dt;
641 	struct lppaca *vpa;
642 	unsigned long stolen;
643 	unsigned long core_stolen;
644 	u64 now;
645 
646 	dt = vcpu->arch.dtl_ptr;
647 	vpa = vcpu->arch.vpa.pinned_addr;
648 	now = mftb();
649 	core_stolen = vcore_stolen_time(vc, now);
650 	stolen = core_stolen - vcpu->arch.stolen_logged;
651 	vcpu->arch.stolen_logged = core_stolen;
652 	spin_lock_irq(&vcpu->arch.tbacct_lock);
653 	stolen += vcpu->arch.busy_stolen;
654 	vcpu->arch.busy_stolen = 0;
655 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
656 	if (!dt || !vpa)
657 		return;
658 	memset(dt, 0, sizeof(struct dtl_entry));
659 	dt->dispatch_reason = 7;
660 	dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
661 	dt->timebase = cpu_to_be64(now + vc->tb_offset);
662 	dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
663 	dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
664 	dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
665 	++dt;
666 	if (dt == vcpu->arch.dtl.pinned_end)
667 		dt = vcpu->arch.dtl.pinned_addr;
668 	vcpu->arch.dtl_ptr = dt;
669 	/* order writing *dt vs. writing vpa->dtl_idx */
670 	smp_wmb();
671 	vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
672 	vcpu->arch.dtl.dirty = true;
673 }
674 
675 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
676 {
677 	if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
678 		return true;
679 	if ((!vcpu->arch.vcore->arch_compat) &&
680 	    cpu_has_feature(CPU_FTR_ARCH_207S))
681 		return true;
682 	return false;
683 }
684 
685 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
686 			     unsigned long resource, unsigned long value1,
687 			     unsigned long value2)
688 {
689 	switch (resource) {
690 	case H_SET_MODE_RESOURCE_SET_CIABR:
691 		if (!kvmppc_power8_compatible(vcpu))
692 			return H_P2;
693 		if (value2)
694 			return H_P4;
695 		if (mflags)
696 			return H_UNSUPPORTED_FLAG_START;
697 		/* Guests can't breakpoint the hypervisor */
698 		if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
699 			return H_P3;
700 		vcpu->arch.ciabr  = value1;
701 		return H_SUCCESS;
702 	case H_SET_MODE_RESOURCE_SET_DAWR:
703 		if (!kvmppc_power8_compatible(vcpu))
704 			return H_P2;
705 		if (mflags)
706 			return H_UNSUPPORTED_FLAG_START;
707 		if (value2 & DABRX_HYP)
708 			return H_P4;
709 		vcpu->arch.dawr  = value1;
710 		vcpu->arch.dawrx = value2;
711 		return H_SUCCESS;
712 	default:
713 		return H_TOO_HARD;
714 	}
715 }
716 
717 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
718 {
719 	struct kvmppc_vcore *vcore = target->arch.vcore;
720 
721 	/*
722 	 * We expect to have been called by the real mode handler
723 	 * (kvmppc_rm_h_confer()) which would have directly returned
724 	 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
725 	 * have useful work to do and should not confer) so we don't
726 	 * recheck that here.
727 	 */
728 
729 	spin_lock(&vcore->lock);
730 	if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
731 	    vcore->vcore_state != VCORE_INACTIVE &&
732 	    vcore->runner)
733 		target = vcore->runner;
734 	spin_unlock(&vcore->lock);
735 
736 	return kvm_vcpu_yield_to(target);
737 }
738 
739 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
740 {
741 	int yield_count = 0;
742 	struct lppaca *lppaca;
743 
744 	spin_lock(&vcpu->arch.vpa_update_lock);
745 	lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
746 	if (lppaca)
747 		yield_count = be32_to_cpu(lppaca->yield_count);
748 	spin_unlock(&vcpu->arch.vpa_update_lock);
749 	return yield_count;
750 }
751 
752 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
753 {
754 	unsigned long req = kvmppc_get_gpr(vcpu, 3);
755 	unsigned long target, ret = H_SUCCESS;
756 	int yield_count;
757 	struct kvm_vcpu *tvcpu;
758 	int idx, rc;
759 
760 	if (req <= MAX_HCALL_OPCODE &&
761 	    !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
762 		return RESUME_HOST;
763 
764 	switch (req) {
765 	case H_CEDE:
766 		break;
767 	case H_PROD:
768 		target = kvmppc_get_gpr(vcpu, 4);
769 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
770 		if (!tvcpu) {
771 			ret = H_PARAMETER;
772 			break;
773 		}
774 		tvcpu->arch.prodded = 1;
775 		smp_mb();
776 		if (vcpu->arch.ceded) {
777 			if (swait_active(&vcpu->wq)) {
778 				swake_up(&vcpu->wq);
779 				vcpu->stat.halt_wakeup++;
780 			}
781 		}
782 		break;
783 	case H_CONFER:
784 		target = kvmppc_get_gpr(vcpu, 4);
785 		if (target == -1)
786 			break;
787 		tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
788 		if (!tvcpu) {
789 			ret = H_PARAMETER;
790 			break;
791 		}
792 		yield_count = kvmppc_get_gpr(vcpu, 5);
793 		if (kvmppc_get_yield_count(tvcpu) != yield_count)
794 			break;
795 		kvm_arch_vcpu_yield_to(tvcpu);
796 		break;
797 	case H_REGISTER_VPA:
798 		ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
799 					kvmppc_get_gpr(vcpu, 5),
800 					kvmppc_get_gpr(vcpu, 6));
801 		break;
802 	case H_RTAS:
803 		if (list_empty(&vcpu->kvm->arch.rtas_tokens))
804 			return RESUME_HOST;
805 
806 		idx = srcu_read_lock(&vcpu->kvm->srcu);
807 		rc = kvmppc_rtas_hcall(vcpu);
808 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
809 
810 		if (rc == -ENOENT)
811 			return RESUME_HOST;
812 		else if (rc == 0)
813 			break;
814 
815 		/* Send the error out to userspace via KVM_RUN */
816 		return rc;
817 	case H_LOGICAL_CI_LOAD:
818 		ret = kvmppc_h_logical_ci_load(vcpu);
819 		if (ret == H_TOO_HARD)
820 			return RESUME_HOST;
821 		break;
822 	case H_LOGICAL_CI_STORE:
823 		ret = kvmppc_h_logical_ci_store(vcpu);
824 		if (ret == H_TOO_HARD)
825 			return RESUME_HOST;
826 		break;
827 	case H_SET_MODE:
828 		ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
829 					kvmppc_get_gpr(vcpu, 5),
830 					kvmppc_get_gpr(vcpu, 6),
831 					kvmppc_get_gpr(vcpu, 7));
832 		if (ret == H_TOO_HARD)
833 			return RESUME_HOST;
834 		break;
835 	case H_XIRR:
836 	case H_CPPR:
837 	case H_EOI:
838 	case H_IPI:
839 	case H_IPOLL:
840 	case H_XIRR_X:
841 		if (kvmppc_xics_enabled(vcpu)) {
842 			ret = kvmppc_xics_hcall(vcpu, req);
843 			break;
844 		}
845 		return RESUME_HOST;
846 	case H_PUT_TCE:
847 		ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
848 						kvmppc_get_gpr(vcpu, 5),
849 						kvmppc_get_gpr(vcpu, 6));
850 		if (ret == H_TOO_HARD)
851 			return RESUME_HOST;
852 		break;
853 	case H_PUT_TCE_INDIRECT:
854 		ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
855 						kvmppc_get_gpr(vcpu, 5),
856 						kvmppc_get_gpr(vcpu, 6),
857 						kvmppc_get_gpr(vcpu, 7));
858 		if (ret == H_TOO_HARD)
859 			return RESUME_HOST;
860 		break;
861 	case H_STUFF_TCE:
862 		ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
863 						kvmppc_get_gpr(vcpu, 5),
864 						kvmppc_get_gpr(vcpu, 6),
865 						kvmppc_get_gpr(vcpu, 7));
866 		if (ret == H_TOO_HARD)
867 			return RESUME_HOST;
868 		break;
869 	default:
870 		return RESUME_HOST;
871 	}
872 	kvmppc_set_gpr(vcpu, 3, ret);
873 	vcpu->arch.hcall_needed = 0;
874 	return RESUME_GUEST;
875 }
876 
877 static int kvmppc_hcall_impl_hv(unsigned long cmd)
878 {
879 	switch (cmd) {
880 	case H_CEDE:
881 	case H_PROD:
882 	case H_CONFER:
883 	case H_REGISTER_VPA:
884 	case H_SET_MODE:
885 	case H_LOGICAL_CI_LOAD:
886 	case H_LOGICAL_CI_STORE:
887 #ifdef CONFIG_KVM_XICS
888 	case H_XIRR:
889 	case H_CPPR:
890 	case H_EOI:
891 	case H_IPI:
892 	case H_IPOLL:
893 	case H_XIRR_X:
894 #endif
895 		return 1;
896 	}
897 
898 	/* See if it's in the real-mode table */
899 	return kvmppc_hcall_impl_hv_realmode(cmd);
900 }
901 
902 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
903 					struct kvm_vcpu *vcpu)
904 {
905 	u32 last_inst;
906 
907 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
908 					EMULATE_DONE) {
909 		/*
910 		 * Fetch failed, so return to guest and
911 		 * try executing it again.
912 		 */
913 		return RESUME_GUEST;
914 	}
915 
916 	if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
917 		run->exit_reason = KVM_EXIT_DEBUG;
918 		run->debug.arch.address = kvmppc_get_pc(vcpu);
919 		return RESUME_HOST;
920 	} else {
921 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
922 		return RESUME_GUEST;
923 	}
924 }
925 
926 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
927 				 struct task_struct *tsk)
928 {
929 	int r = RESUME_HOST;
930 
931 	vcpu->stat.sum_exits++;
932 
933 	/*
934 	 * This can happen if an interrupt occurs in the last stages
935 	 * of guest entry or the first stages of guest exit (i.e. after
936 	 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
937 	 * and before setting it to KVM_GUEST_MODE_HOST_HV).
938 	 * That can happen due to a bug, or due to a machine check
939 	 * occurring at just the wrong time.
940 	 */
941 	if (vcpu->arch.shregs.msr & MSR_HV) {
942 		printk(KERN_EMERG "KVM trap in HV mode!\n");
943 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
944 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
945 			vcpu->arch.shregs.msr);
946 		kvmppc_dump_regs(vcpu);
947 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
948 		run->hw.hardware_exit_reason = vcpu->arch.trap;
949 		return RESUME_HOST;
950 	}
951 	run->exit_reason = KVM_EXIT_UNKNOWN;
952 	run->ready_for_interrupt_injection = 1;
953 	switch (vcpu->arch.trap) {
954 	/* We're good on these - the host merely wanted to get our attention */
955 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
956 		vcpu->stat.dec_exits++;
957 		r = RESUME_GUEST;
958 		break;
959 	case BOOK3S_INTERRUPT_EXTERNAL:
960 	case BOOK3S_INTERRUPT_H_DOORBELL:
961 	case BOOK3S_INTERRUPT_H_VIRT:
962 		vcpu->stat.ext_intr_exits++;
963 		r = RESUME_GUEST;
964 		break;
965 	/* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
966 	case BOOK3S_INTERRUPT_HMI:
967 	case BOOK3S_INTERRUPT_PERFMON:
968 		r = RESUME_GUEST;
969 		break;
970 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
971 		/*
972 		 * Deliver a machine check interrupt to the guest.
973 		 * We have to do this, even if the host has handled the
974 		 * machine check, because machine checks use SRR0/1 and
975 		 * the interrupt might have trashed guest state in them.
976 		 */
977 		kvmppc_book3s_queue_irqprio(vcpu,
978 					    BOOK3S_INTERRUPT_MACHINE_CHECK);
979 		r = RESUME_GUEST;
980 		break;
981 	case BOOK3S_INTERRUPT_PROGRAM:
982 	{
983 		ulong flags;
984 		/*
985 		 * Normally program interrupts are delivered directly
986 		 * to the guest by the hardware, but we can get here
987 		 * as a result of a hypervisor emulation interrupt
988 		 * (e40) getting turned into a 700 by BML RTAS.
989 		 */
990 		flags = vcpu->arch.shregs.msr & 0x1f0000ull;
991 		kvmppc_core_queue_program(vcpu, flags);
992 		r = RESUME_GUEST;
993 		break;
994 	}
995 	case BOOK3S_INTERRUPT_SYSCALL:
996 	{
997 		/* hcall - punt to userspace */
998 		int i;
999 
1000 		/* hypercall with MSR_PR has already been handled in rmode,
1001 		 * and never reaches here.
1002 		 */
1003 
1004 		run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1005 		for (i = 0; i < 9; ++i)
1006 			run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1007 		run->exit_reason = KVM_EXIT_PAPR_HCALL;
1008 		vcpu->arch.hcall_needed = 1;
1009 		r = RESUME_HOST;
1010 		break;
1011 	}
1012 	/*
1013 	 * We get these next two if the guest accesses a page which it thinks
1014 	 * it has mapped but which is not actually present, either because
1015 	 * it is for an emulated I/O device or because the corresonding
1016 	 * host page has been paged out.  Any other HDSI/HISI interrupts
1017 	 * have been handled already.
1018 	 */
1019 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1020 		r = RESUME_PAGE_FAULT;
1021 		break;
1022 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
1023 		vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1024 		vcpu->arch.fault_dsisr = 0;
1025 		r = RESUME_PAGE_FAULT;
1026 		break;
1027 	/*
1028 	 * This occurs if the guest executes an illegal instruction.
1029 	 * If the guest debug is disabled, generate a program interrupt
1030 	 * to the guest. If guest debug is enabled, we need to check
1031 	 * whether the instruction is a software breakpoint instruction.
1032 	 * Accordingly return to Guest or Host.
1033 	 */
1034 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1035 		if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1036 			vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1037 				swab32(vcpu->arch.emul_inst) :
1038 				vcpu->arch.emul_inst;
1039 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1040 			r = kvmppc_emulate_debug_inst(run, vcpu);
1041 		} else {
1042 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1043 			r = RESUME_GUEST;
1044 		}
1045 		break;
1046 	/*
1047 	 * This occurs if the guest (kernel or userspace), does something that
1048 	 * is prohibited by HFSCR.  We just generate a program interrupt to
1049 	 * the guest.
1050 	 */
1051 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1052 		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1053 		r = RESUME_GUEST;
1054 		break;
1055 	case BOOK3S_INTERRUPT_HV_RM_HARD:
1056 		r = RESUME_PASSTHROUGH;
1057 		break;
1058 	default:
1059 		kvmppc_dump_regs(vcpu);
1060 		printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1061 			vcpu->arch.trap, kvmppc_get_pc(vcpu),
1062 			vcpu->arch.shregs.msr);
1063 		run->hw.hardware_exit_reason = vcpu->arch.trap;
1064 		r = RESUME_HOST;
1065 		break;
1066 	}
1067 
1068 	return r;
1069 }
1070 
1071 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1072 					    struct kvm_sregs *sregs)
1073 {
1074 	int i;
1075 
1076 	memset(sregs, 0, sizeof(struct kvm_sregs));
1077 	sregs->pvr = vcpu->arch.pvr;
1078 	for (i = 0; i < vcpu->arch.slb_max; i++) {
1079 		sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1080 		sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1081 	}
1082 
1083 	return 0;
1084 }
1085 
1086 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1087 					    struct kvm_sregs *sregs)
1088 {
1089 	int i, j;
1090 
1091 	/* Only accept the same PVR as the host's, since we can't spoof it */
1092 	if (sregs->pvr != vcpu->arch.pvr)
1093 		return -EINVAL;
1094 
1095 	j = 0;
1096 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
1097 		if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1098 			vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1099 			vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1100 			++j;
1101 		}
1102 	}
1103 	vcpu->arch.slb_max = j;
1104 
1105 	return 0;
1106 }
1107 
1108 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1109 		bool preserve_top32)
1110 {
1111 	struct kvm *kvm = vcpu->kvm;
1112 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
1113 	u64 mask;
1114 
1115 	mutex_lock(&kvm->lock);
1116 	spin_lock(&vc->lock);
1117 	/*
1118 	 * If ILE (interrupt little-endian) has changed, update the
1119 	 * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1120 	 */
1121 	if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1122 		struct kvm_vcpu *vcpu;
1123 		int i;
1124 
1125 		kvm_for_each_vcpu(i, vcpu, kvm) {
1126 			if (vcpu->arch.vcore != vc)
1127 				continue;
1128 			if (new_lpcr & LPCR_ILE)
1129 				vcpu->arch.intr_msr |= MSR_LE;
1130 			else
1131 				vcpu->arch.intr_msr &= ~MSR_LE;
1132 		}
1133 	}
1134 
1135 	/*
1136 	 * Userspace can only modify DPFD (default prefetch depth),
1137 	 * ILE (interrupt little-endian) and TC (translation control).
1138 	 * On POWER8 userspace can also modify AIL (alt. interrupt loc.)
1139 	 */
1140 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1141 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
1142 		mask |= LPCR_AIL;
1143 
1144 	/* Broken 32-bit version of LPCR must not clear top bits */
1145 	if (preserve_top32)
1146 		mask &= 0xFFFFFFFF;
1147 	vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1148 	spin_unlock(&vc->lock);
1149 	mutex_unlock(&kvm->lock);
1150 }
1151 
1152 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1153 				 union kvmppc_one_reg *val)
1154 {
1155 	int r = 0;
1156 	long int i;
1157 
1158 	switch (id) {
1159 	case KVM_REG_PPC_DEBUG_INST:
1160 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1161 		break;
1162 	case KVM_REG_PPC_HIOR:
1163 		*val = get_reg_val(id, 0);
1164 		break;
1165 	case KVM_REG_PPC_DABR:
1166 		*val = get_reg_val(id, vcpu->arch.dabr);
1167 		break;
1168 	case KVM_REG_PPC_DABRX:
1169 		*val = get_reg_val(id, vcpu->arch.dabrx);
1170 		break;
1171 	case KVM_REG_PPC_DSCR:
1172 		*val = get_reg_val(id, vcpu->arch.dscr);
1173 		break;
1174 	case KVM_REG_PPC_PURR:
1175 		*val = get_reg_val(id, vcpu->arch.purr);
1176 		break;
1177 	case KVM_REG_PPC_SPURR:
1178 		*val = get_reg_val(id, vcpu->arch.spurr);
1179 		break;
1180 	case KVM_REG_PPC_AMR:
1181 		*val = get_reg_val(id, vcpu->arch.amr);
1182 		break;
1183 	case KVM_REG_PPC_UAMOR:
1184 		*val = get_reg_val(id, vcpu->arch.uamor);
1185 		break;
1186 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1187 		i = id - KVM_REG_PPC_MMCR0;
1188 		*val = get_reg_val(id, vcpu->arch.mmcr[i]);
1189 		break;
1190 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1191 		i = id - KVM_REG_PPC_PMC1;
1192 		*val = get_reg_val(id, vcpu->arch.pmc[i]);
1193 		break;
1194 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1195 		i = id - KVM_REG_PPC_SPMC1;
1196 		*val = get_reg_val(id, vcpu->arch.spmc[i]);
1197 		break;
1198 	case KVM_REG_PPC_SIAR:
1199 		*val = get_reg_val(id, vcpu->arch.siar);
1200 		break;
1201 	case KVM_REG_PPC_SDAR:
1202 		*val = get_reg_val(id, vcpu->arch.sdar);
1203 		break;
1204 	case KVM_REG_PPC_SIER:
1205 		*val = get_reg_val(id, vcpu->arch.sier);
1206 		break;
1207 	case KVM_REG_PPC_IAMR:
1208 		*val = get_reg_val(id, vcpu->arch.iamr);
1209 		break;
1210 	case KVM_REG_PPC_PSPB:
1211 		*val = get_reg_val(id, vcpu->arch.pspb);
1212 		break;
1213 	case KVM_REG_PPC_DPDES:
1214 		*val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1215 		break;
1216 	case KVM_REG_PPC_VTB:
1217 		*val = get_reg_val(id, vcpu->arch.vcore->vtb);
1218 		break;
1219 	case KVM_REG_PPC_DAWR:
1220 		*val = get_reg_val(id, vcpu->arch.dawr);
1221 		break;
1222 	case KVM_REG_PPC_DAWRX:
1223 		*val = get_reg_val(id, vcpu->arch.dawrx);
1224 		break;
1225 	case KVM_REG_PPC_CIABR:
1226 		*val = get_reg_val(id, vcpu->arch.ciabr);
1227 		break;
1228 	case KVM_REG_PPC_CSIGR:
1229 		*val = get_reg_val(id, vcpu->arch.csigr);
1230 		break;
1231 	case KVM_REG_PPC_TACR:
1232 		*val = get_reg_val(id, vcpu->arch.tacr);
1233 		break;
1234 	case KVM_REG_PPC_TCSCR:
1235 		*val = get_reg_val(id, vcpu->arch.tcscr);
1236 		break;
1237 	case KVM_REG_PPC_PID:
1238 		*val = get_reg_val(id, vcpu->arch.pid);
1239 		break;
1240 	case KVM_REG_PPC_ACOP:
1241 		*val = get_reg_val(id, vcpu->arch.acop);
1242 		break;
1243 	case KVM_REG_PPC_WORT:
1244 		*val = get_reg_val(id, vcpu->arch.wort);
1245 		break;
1246 	case KVM_REG_PPC_TIDR:
1247 		*val = get_reg_val(id, vcpu->arch.tid);
1248 		break;
1249 	case KVM_REG_PPC_PSSCR:
1250 		*val = get_reg_val(id, vcpu->arch.psscr);
1251 		break;
1252 	case KVM_REG_PPC_VPA_ADDR:
1253 		spin_lock(&vcpu->arch.vpa_update_lock);
1254 		*val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1255 		spin_unlock(&vcpu->arch.vpa_update_lock);
1256 		break;
1257 	case KVM_REG_PPC_VPA_SLB:
1258 		spin_lock(&vcpu->arch.vpa_update_lock);
1259 		val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1260 		val->vpaval.length = vcpu->arch.slb_shadow.len;
1261 		spin_unlock(&vcpu->arch.vpa_update_lock);
1262 		break;
1263 	case KVM_REG_PPC_VPA_DTL:
1264 		spin_lock(&vcpu->arch.vpa_update_lock);
1265 		val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1266 		val->vpaval.length = vcpu->arch.dtl.len;
1267 		spin_unlock(&vcpu->arch.vpa_update_lock);
1268 		break;
1269 	case KVM_REG_PPC_TB_OFFSET:
1270 		*val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1271 		break;
1272 	case KVM_REG_PPC_LPCR:
1273 	case KVM_REG_PPC_LPCR_64:
1274 		*val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1275 		break;
1276 	case KVM_REG_PPC_PPR:
1277 		*val = get_reg_val(id, vcpu->arch.ppr);
1278 		break;
1279 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1280 	case KVM_REG_PPC_TFHAR:
1281 		*val = get_reg_val(id, vcpu->arch.tfhar);
1282 		break;
1283 	case KVM_REG_PPC_TFIAR:
1284 		*val = get_reg_val(id, vcpu->arch.tfiar);
1285 		break;
1286 	case KVM_REG_PPC_TEXASR:
1287 		*val = get_reg_val(id, vcpu->arch.texasr);
1288 		break;
1289 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1290 		i = id - KVM_REG_PPC_TM_GPR0;
1291 		*val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1292 		break;
1293 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1294 	{
1295 		int j;
1296 		i = id - KVM_REG_PPC_TM_VSR0;
1297 		if (i < 32)
1298 			for (j = 0; j < TS_FPRWIDTH; j++)
1299 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1300 		else {
1301 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1302 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1303 			else
1304 				r = -ENXIO;
1305 		}
1306 		break;
1307 	}
1308 	case KVM_REG_PPC_TM_CR:
1309 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1310 		break;
1311 	case KVM_REG_PPC_TM_XER:
1312 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1313 		break;
1314 	case KVM_REG_PPC_TM_LR:
1315 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1316 		break;
1317 	case KVM_REG_PPC_TM_CTR:
1318 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1319 		break;
1320 	case KVM_REG_PPC_TM_FPSCR:
1321 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1322 		break;
1323 	case KVM_REG_PPC_TM_AMR:
1324 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1325 		break;
1326 	case KVM_REG_PPC_TM_PPR:
1327 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1328 		break;
1329 	case KVM_REG_PPC_TM_VRSAVE:
1330 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1331 		break;
1332 	case KVM_REG_PPC_TM_VSCR:
1333 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1334 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1335 		else
1336 			r = -ENXIO;
1337 		break;
1338 	case KVM_REG_PPC_TM_DSCR:
1339 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1340 		break;
1341 	case KVM_REG_PPC_TM_TAR:
1342 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1343 		break;
1344 #endif
1345 	case KVM_REG_PPC_ARCH_COMPAT:
1346 		*val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1347 		break;
1348 	default:
1349 		r = -EINVAL;
1350 		break;
1351 	}
1352 
1353 	return r;
1354 }
1355 
1356 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1357 				 union kvmppc_one_reg *val)
1358 {
1359 	int r = 0;
1360 	long int i;
1361 	unsigned long addr, len;
1362 
1363 	switch (id) {
1364 	case KVM_REG_PPC_HIOR:
1365 		/* Only allow this to be set to zero */
1366 		if (set_reg_val(id, *val))
1367 			r = -EINVAL;
1368 		break;
1369 	case KVM_REG_PPC_DABR:
1370 		vcpu->arch.dabr = set_reg_val(id, *val);
1371 		break;
1372 	case KVM_REG_PPC_DABRX:
1373 		vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1374 		break;
1375 	case KVM_REG_PPC_DSCR:
1376 		vcpu->arch.dscr = set_reg_val(id, *val);
1377 		break;
1378 	case KVM_REG_PPC_PURR:
1379 		vcpu->arch.purr = set_reg_val(id, *val);
1380 		break;
1381 	case KVM_REG_PPC_SPURR:
1382 		vcpu->arch.spurr = set_reg_val(id, *val);
1383 		break;
1384 	case KVM_REG_PPC_AMR:
1385 		vcpu->arch.amr = set_reg_val(id, *val);
1386 		break;
1387 	case KVM_REG_PPC_UAMOR:
1388 		vcpu->arch.uamor = set_reg_val(id, *val);
1389 		break;
1390 	case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1391 		i = id - KVM_REG_PPC_MMCR0;
1392 		vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1393 		break;
1394 	case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1395 		i = id - KVM_REG_PPC_PMC1;
1396 		vcpu->arch.pmc[i] = set_reg_val(id, *val);
1397 		break;
1398 	case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1399 		i = id - KVM_REG_PPC_SPMC1;
1400 		vcpu->arch.spmc[i] = set_reg_val(id, *val);
1401 		break;
1402 	case KVM_REG_PPC_SIAR:
1403 		vcpu->arch.siar = set_reg_val(id, *val);
1404 		break;
1405 	case KVM_REG_PPC_SDAR:
1406 		vcpu->arch.sdar = set_reg_val(id, *val);
1407 		break;
1408 	case KVM_REG_PPC_SIER:
1409 		vcpu->arch.sier = set_reg_val(id, *val);
1410 		break;
1411 	case KVM_REG_PPC_IAMR:
1412 		vcpu->arch.iamr = set_reg_val(id, *val);
1413 		break;
1414 	case KVM_REG_PPC_PSPB:
1415 		vcpu->arch.pspb = set_reg_val(id, *val);
1416 		break;
1417 	case KVM_REG_PPC_DPDES:
1418 		vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1419 		break;
1420 	case KVM_REG_PPC_VTB:
1421 		vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1422 		break;
1423 	case KVM_REG_PPC_DAWR:
1424 		vcpu->arch.dawr = set_reg_val(id, *val);
1425 		break;
1426 	case KVM_REG_PPC_DAWRX:
1427 		vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1428 		break;
1429 	case KVM_REG_PPC_CIABR:
1430 		vcpu->arch.ciabr = set_reg_val(id, *val);
1431 		/* Don't allow setting breakpoints in hypervisor code */
1432 		if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1433 			vcpu->arch.ciabr &= ~CIABR_PRIV;	/* disable */
1434 		break;
1435 	case KVM_REG_PPC_CSIGR:
1436 		vcpu->arch.csigr = set_reg_val(id, *val);
1437 		break;
1438 	case KVM_REG_PPC_TACR:
1439 		vcpu->arch.tacr = set_reg_val(id, *val);
1440 		break;
1441 	case KVM_REG_PPC_TCSCR:
1442 		vcpu->arch.tcscr = set_reg_val(id, *val);
1443 		break;
1444 	case KVM_REG_PPC_PID:
1445 		vcpu->arch.pid = set_reg_val(id, *val);
1446 		break;
1447 	case KVM_REG_PPC_ACOP:
1448 		vcpu->arch.acop = set_reg_val(id, *val);
1449 		break;
1450 	case KVM_REG_PPC_WORT:
1451 		vcpu->arch.wort = set_reg_val(id, *val);
1452 		break;
1453 	case KVM_REG_PPC_TIDR:
1454 		vcpu->arch.tid = set_reg_val(id, *val);
1455 		break;
1456 	case KVM_REG_PPC_PSSCR:
1457 		vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1458 		break;
1459 	case KVM_REG_PPC_VPA_ADDR:
1460 		addr = set_reg_val(id, *val);
1461 		r = -EINVAL;
1462 		if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1463 			      vcpu->arch.dtl.next_gpa))
1464 			break;
1465 		r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1466 		break;
1467 	case KVM_REG_PPC_VPA_SLB:
1468 		addr = val->vpaval.addr;
1469 		len = val->vpaval.length;
1470 		r = -EINVAL;
1471 		if (addr && !vcpu->arch.vpa.next_gpa)
1472 			break;
1473 		r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1474 		break;
1475 	case KVM_REG_PPC_VPA_DTL:
1476 		addr = val->vpaval.addr;
1477 		len = val->vpaval.length;
1478 		r = -EINVAL;
1479 		if (addr && (len < sizeof(struct dtl_entry) ||
1480 			     !vcpu->arch.vpa.next_gpa))
1481 			break;
1482 		len -= len % sizeof(struct dtl_entry);
1483 		r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1484 		break;
1485 	case KVM_REG_PPC_TB_OFFSET:
1486 		/* round up to multiple of 2^24 */
1487 		vcpu->arch.vcore->tb_offset =
1488 			ALIGN(set_reg_val(id, *val), 1UL << 24);
1489 		break;
1490 	case KVM_REG_PPC_LPCR:
1491 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1492 		break;
1493 	case KVM_REG_PPC_LPCR_64:
1494 		kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1495 		break;
1496 	case KVM_REG_PPC_PPR:
1497 		vcpu->arch.ppr = set_reg_val(id, *val);
1498 		break;
1499 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1500 	case KVM_REG_PPC_TFHAR:
1501 		vcpu->arch.tfhar = set_reg_val(id, *val);
1502 		break;
1503 	case KVM_REG_PPC_TFIAR:
1504 		vcpu->arch.tfiar = set_reg_val(id, *val);
1505 		break;
1506 	case KVM_REG_PPC_TEXASR:
1507 		vcpu->arch.texasr = set_reg_val(id, *val);
1508 		break;
1509 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1510 		i = id - KVM_REG_PPC_TM_GPR0;
1511 		vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1512 		break;
1513 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1514 	{
1515 		int j;
1516 		i = id - KVM_REG_PPC_TM_VSR0;
1517 		if (i < 32)
1518 			for (j = 0; j < TS_FPRWIDTH; j++)
1519 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1520 		else
1521 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1522 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1523 			else
1524 				r = -ENXIO;
1525 		break;
1526 	}
1527 	case KVM_REG_PPC_TM_CR:
1528 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1529 		break;
1530 	case KVM_REG_PPC_TM_XER:
1531 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1532 		break;
1533 	case KVM_REG_PPC_TM_LR:
1534 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1535 		break;
1536 	case KVM_REG_PPC_TM_CTR:
1537 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1538 		break;
1539 	case KVM_REG_PPC_TM_FPSCR:
1540 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1541 		break;
1542 	case KVM_REG_PPC_TM_AMR:
1543 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1544 		break;
1545 	case KVM_REG_PPC_TM_PPR:
1546 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1547 		break;
1548 	case KVM_REG_PPC_TM_VRSAVE:
1549 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1550 		break;
1551 	case KVM_REG_PPC_TM_VSCR:
1552 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1553 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1554 		else
1555 			r = - ENXIO;
1556 		break;
1557 	case KVM_REG_PPC_TM_DSCR:
1558 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1559 		break;
1560 	case KVM_REG_PPC_TM_TAR:
1561 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1562 		break;
1563 #endif
1564 	case KVM_REG_PPC_ARCH_COMPAT:
1565 		r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1566 		break;
1567 	default:
1568 		r = -EINVAL;
1569 		break;
1570 	}
1571 
1572 	return r;
1573 }
1574 
1575 /*
1576  * On POWER9, threads are independent and can be in different partitions.
1577  * Therefore we consider each thread to be a subcore.
1578  * There is a restriction that all threads have to be in the same
1579  * MMU mode (radix or HPT), unfortunately, but since we only support
1580  * HPT guests on a HPT host so far, that isn't an impediment yet.
1581  */
1582 static int threads_per_vcore(void)
1583 {
1584 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1585 		return 1;
1586 	return threads_per_subcore;
1587 }
1588 
1589 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1590 {
1591 	struct kvmppc_vcore *vcore;
1592 
1593 	vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1594 
1595 	if (vcore == NULL)
1596 		return NULL;
1597 
1598 	spin_lock_init(&vcore->lock);
1599 	spin_lock_init(&vcore->stoltb_lock);
1600 	init_swait_queue_head(&vcore->wq);
1601 	vcore->preempt_tb = TB_NIL;
1602 	vcore->lpcr = kvm->arch.lpcr;
1603 	vcore->first_vcpuid = core * threads_per_vcore();
1604 	vcore->kvm = kvm;
1605 	INIT_LIST_HEAD(&vcore->preempt_list);
1606 
1607 	return vcore;
1608 }
1609 
1610 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1611 static struct debugfs_timings_element {
1612 	const char *name;
1613 	size_t offset;
1614 } timings[] = {
1615 	{"rm_entry",	offsetof(struct kvm_vcpu, arch.rm_entry)},
1616 	{"rm_intr",	offsetof(struct kvm_vcpu, arch.rm_intr)},
1617 	{"rm_exit",	offsetof(struct kvm_vcpu, arch.rm_exit)},
1618 	{"guest",	offsetof(struct kvm_vcpu, arch.guest_time)},
1619 	{"cede",	offsetof(struct kvm_vcpu, arch.cede_time)},
1620 };
1621 
1622 #define N_TIMINGS	(sizeof(timings) / sizeof(timings[0]))
1623 
1624 struct debugfs_timings_state {
1625 	struct kvm_vcpu	*vcpu;
1626 	unsigned int	buflen;
1627 	char		buf[N_TIMINGS * 100];
1628 };
1629 
1630 static int debugfs_timings_open(struct inode *inode, struct file *file)
1631 {
1632 	struct kvm_vcpu *vcpu = inode->i_private;
1633 	struct debugfs_timings_state *p;
1634 
1635 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1636 	if (!p)
1637 		return -ENOMEM;
1638 
1639 	kvm_get_kvm(vcpu->kvm);
1640 	p->vcpu = vcpu;
1641 	file->private_data = p;
1642 
1643 	return nonseekable_open(inode, file);
1644 }
1645 
1646 static int debugfs_timings_release(struct inode *inode, struct file *file)
1647 {
1648 	struct debugfs_timings_state *p = file->private_data;
1649 
1650 	kvm_put_kvm(p->vcpu->kvm);
1651 	kfree(p);
1652 	return 0;
1653 }
1654 
1655 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1656 				    size_t len, loff_t *ppos)
1657 {
1658 	struct debugfs_timings_state *p = file->private_data;
1659 	struct kvm_vcpu *vcpu = p->vcpu;
1660 	char *s, *buf_end;
1661 	struct kvmhv_tb_accumulator tb;
1662 	u64 count;
1663 	loff_t pos;
1664 	ssize_t n;
1665 	int i, loops;
1666 	bool ok;
1667 
1668 	if (!p->buflen) {
1669 		s = p->buf;
1670 		buf_end = s + sizeof(p->buf);
1671 		for (i = 0; i < N_TIMINGS; ++i) {
1672 			struct kvmhv_tb_accumulator *acc;
1673 
1674 			acc = (struct kvmhv_tb_accumulator *)
1675 				((unsigned long)vcpu + timings[i].offset);
1676 			ok = false;
1677 			for (loops = 0; loops < 1000; ++loops) {
1678 				count = acc->seqcount;
1679 				if (!(count & 1)) {
1680 					smp_rmb();
1681 					tb = *acc;
1682 					smp_rmb();
1683 					if (count == acc->seqcount) {
1684 						ok = true;
1685 						break;
1686 					}
1687 				}
1688 				udelay(1);
1689 			}
1690 			if (!ok)
1691 				snprintf(s, buf_end - s, "%s: stuck\n",
1692 					timings[i].name);
1693 			else
1694 				snprintf(s, buf_end - s,
1695 					"%s: %llu %llu %llu %llu\n",
1696 					timings[i].name, count / 2,
1697 					tb_to_ns(tb.tb_total),
1698 					tb_to_ns(tb.tb_min),
1699 					tb_to_ns(tb.tb_max));
1700 			s += strlen(s);
1701 		}
1702 		p->buflen = s - p->buf;
1703 	}
1704 
1705 	pos = *ppos;
1706 	if (pos >= p->buflen)
1707 		return 0;
1708 	if (len > p->buflen - pos)
1709 		len = p->buflen - pos;
1710 	n = copy_to_user(buf, p->buf + pos, len);
1711 	if (n) {
1712 		if (n == len)
1713 			return -EFAULT;
1714 		len -= n;
1715 	}
1716 	*ppos = pos + len;
1717 	return len;
1718 }
1719 
1720 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1721 				     size_t len, loff_t *ppos)
1722 {
1723 	return -EACCES;
1724 }
1725 
1726 static const struct file_operations debugfs_timings_ops = {
1727 	.owner	 = THIS_MODULE,
1728 	.open	 = debugfs_timings_open,
1729 	.release = debugfs_timings_release,
1730 	.read	 = debugfs_timings_read,
1731 	.write	 = debugfs_timings_write,
1732 	.llseek	 = generic_file_llseek,
1733 };
1734 
1735 /* Create a debugfs directory for the vcpu */
1736 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1737 {
1738 	char buf[16];
1739 	struct kvm *kvm = vcpu->kvm;
1740 
1741 	snprintf(buf, sizeof(buf), "vcpu%u", id);
1742 	if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1743 		return;
1744 	vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1745 	if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1746 		return;
1747 	vcpu->arch.debugfs_timings =
1748 		debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1749 				    vcpu, &debugfs_timings_ops);
1750 }
1751 
1752 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1753 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1754 {
1755 }
1756 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1757 
1758 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1759 						   unsigned int id)
1760 {
1761 	struct kvm_vcpu *vcpu;
1762 	int err = -EINVAL;
1763 	int core;
1764 	struct kvmppc_vcore *vcore;
1765 
1766 	core = id / threads_per_vcore();
1767 	if (core >= KVM_MAX_VCORES)
1768 		goto out;
1769 
1770 	err = -ENOMEM;
1771 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1772 	if (!vcpu)
1773 		goto out;
1774 
1775 	err = kvm_vcpu_init(vcpu, kvm, id);
1776 	if (err)
1777 		goto free_vcpu;
1778 
1779 	vcpu->arch.shared = &vcpu->arch.shregs;
1780 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1781 	/*
1782 	 * The shared struct is never shared on HV,
1783 	 * so we can always use host endianness
1784 	 */
1785 #ifdef __BIG_ENDIAN__
1786 	vcpu->arch.shared_big_endian = true;
1787 #else
1788 	vcpu->arch.shared_big_endian = false;
1789 #endif
1790 #endif
1791 	vcpu->arch.mmcr[0] = MMCR0_FC;
1792 	vcpu->arch.ctrl = CTRL_RUNLATCH;
1793 	/* default to host PVR, since we can't spoof it */
1794 	kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1795 	spin_lock_init(&vcpu->arch.vpa_update_lock);
1796 	spin_lock_init(&vcpu->arch.tbacct_lock);
1797 	vcpu->arch.busy_preempt = TB_NIL;
1798 	vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1799 
1800 	kvmppc_mmu_book3s_hv_init(vcpu);
1801 
1802 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1803 
1804 	init_waitqueue_head(&vcpu->arch.cpu_run);
1805 
1806 	mutex_lock(&kvm->lock);
1807 	vcore = kvm->arch.vcores[core];
1808 	if (!vcore) {
1809 		vcore = kvmppc_vcore_create(kvm, core);
1810 		kvm->arch.vcores[core] = vcore;
1811 		kvm->arch.online_vcores++;
1812 	}
1813 	mutex_unlock(&kvm->lock);
1814 
1815 	if (!vcore)
1816 		goto free_vcpu;
1817 
1818 	spin_lock(&vcore->lock);
1819 	++vcore->num_threads;
1820 	spin_unlock(&vcore->lock);
1821 	vcpu->arch.vcore = vcore;
1822 	vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1823 	vcpu->arch.thread_cpu = -1;
1824 
1825 	vcpu->arch.cpu_type = KVM_CPU_3S_64;
1826 	kvmppc_sanity_check(vcpu);
1827 
1828 	debugfs_vcpu_init(vcpu, id);
1829 
1830 	return vcpu;
1831 
1832 free_vcpu:
1833 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1834 out:
1835 	return ERR_PTR(err);
1836 }
1837 
1838 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1839 {
1840 	if (vpa->pinned_addr)
1841 		kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1842 					vpa->dirty);
1843 }
1844 
1845 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1846 {
1847 	spin_lock(&vcpu->arch.vpa_update_lock);
1848 	unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1849 	unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1850 	unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1851 	spin_unlock(&vcpu->arch.vpa_update_lock);
1852 	kvm_vcpu_uninit(vcpu);
1853 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1854 }
1855 
1856 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1857 {
1858 	/* Indicate we want to get back into the guest */
1859 	return 1;
1860 }
1861 
1862 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1863 {
1864 	unsigned long dec_nsec, now;
1865 
1866 	now = get_tb();
1867 	if (now > vcpu->arch.dec_expires) {
1868 		/* decrementer has already gone negative */
1869 		kvmppc_core_queue_dec(vcpu);
1870 		kvmppc_core_prepare_to_enter(vcpu);
1871 		return;
1872 	}
1873 	dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1874 		   / tb_ticks_per_sec;
1875 	hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
1876 	vcpu->arch.timer_running = 1;
1877 }
1878 
1879 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1880 {
1881 	vcpu->arch.ceded = 0;
1882 	if (vcpu->arch.timer_running) {
1883 		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1884 		vcpu->arch.timer_running = 0;
1885 	}
1886 }
1887 
1888 extern void __kvmppc_vcore_entry(void);
1889 
1890 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1891 				   struct kvm_vcpu *vcpu)
1892 {
1893 	u64 now;
1894 
1895 	if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1896 		return;
1897 	spin_lock_irq(&vcpu->arch.tbacct_lock);
1898 	now = mftb();
1899 	vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1900 		vcpu->arch.stolen_logged;
1901 	vcpu->arch.busy_preempt = now;
1902 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1903 	spin_unlock_irq(&vcpu->arch.tbacct_lock);
1904 	--vc->n_runnable;
1905 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1906 }
1907 
1908 static int kvmppc_grab_hwthread(int cpu)
1909 {
1910 	struct paca_struct *tpaca;
1911 	long timeout = 10000;
1912 
1913 	tpaca = &paca[cpu];
1914 
1915 	/* Ensure the thread won't go into the kernel if it wakes */
1916 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1917 	tpaca->kvm_hstate.kvm_vcore = NULL;
1918 	tpaca->kvm_hstate.napping = 0;
1919 	smp_wmb();
1920 	tpaca->kvm_hstate.hwthread_req = 1;
1921 
1922 	/*
1923 	 * If the thread is already executing in the kernel (e.g. handling
1924 	 * a stray interrupt), wait for it to get back to nap mode.
1925 	 * The smp_mb() is to ensure that our setting of hwthread_req
1926 	 * is visible before we look at hwthread_state, so if this
1927 	 * races with the code at system_reset_pSeries and the thread
1928 	 * misses our setting of hwthread_req, we are sure to see its
1929 	 * setting of hwthread_state, and vice versa.
1930 	 */
1931 	smp_mb();
1932 	while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1933 		if (--timeout <= 0) {
1934 			pr_err("KVM: couldn't grab cpu %d\n", cpu);
1935 			return -EBUSY;
1936 		}
1937 		udelay(1);
1938 	}
1939 	return 0;
1940 }
1941 
1942 static void kvmppc_release_hwthread(int cpu)
1943 {
1944 	struct paca_struct *tpaca;
1945 
1946 	tpaca = &paca[cpu];
1947 	tpaca->kvm_hstate.hwthread_req = 0;
1948 	tpaca->kvm_hstate.kvm_vcpu = NULL;
1949 	tpaca->kvm_hstate.kvm_vcore = NULL;
1950 	tpaca->kvm_hstate.kvm_split_mode = NULL;
1951 }
1952 
1953 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1954 {
1955 	int cpu;
1956 	struct paca_struct *tpaca;
1957 	struct kvmppc_vcore *mvc = vc->master_vcore;
1958 
1959 	cpu = vc->pcpu;
1960 	if (vcpu) {
1961 		if (vcpu->arch.timer_running) {
1962 			hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1963 			vcpu->arch.timer_running = 0;
1964 		}
1965 		cpu += vcpu->arch.ptid;
1966 		vcpu->cpu = mvc->pcpu;
1967 		vcpu->arch.thread_cpu = cpu;
1968 	}
1969 	tpaca = &paca[cpu];
1970 	tpaca->kvm_hstate.kvm_vcpu = vcpu;
1971 	tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
1972 	/* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
1973 	smp_wmb();
1974 	tpaca->kvm_hstate.kvm_vcore = mvc;
1975 	if (cpu != smp_processor_id())
1976 		kvmppc_ipi_thread(cpu);
1977 }
1978 
1979 static void kvmppc_wait_for_nap(void)
1980 {
1981 	int cpu = smp_processor_id();
1982 	int i, loops;
1983 	int n_threads = threads_per_vcore();
1984 
1985 	if (n_threads <= 1)
1986 		return;
1987 	for (loops = 0; loops < 1000000; ++loops) {
1988 		/*
1989 		 * Check if all threads are finished.
1990 		 * We set the vcore pointer when starting a thread
1991 		 * and the thread clears it when finished, so we look
1992 		 * for any threads that still have a non-NULL vcore ptr.
1993 		 */
1994 		for (i = 1; i < n_threads; ++i)
1995 			if (paca[cpu + i].kvm_hstate.kvm_vcore)
1996 				break;
1997 		if (i == n_threads) {
1998 			HMT_medium();
1999 			return;
2000 		}
2001 		HMT_low();
2002 	}
2003 	HMT_medium();
2004 	for (i = 1; i < n_threads; ++i)
2005 		if (paca[cpu + i].kvm_hstate.kvm_vcore)
2006 			pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2007 }
2008 
2009 /*
2010  * Check that we are on thread 0 and that any other threads in
2011  * this core are off-line.  Then grab the threads so they can't
2012  * enter the kernel.
2013  */
2014 static int on_primary_thread(void)
2015 {
2016 	int cpu = smp_processor_id();
2017 	int thr;
2018 
2019 	/* Are we on a primary subcore? */
2020 	if (cpu_thread_in_subcore(cpu))
2021 		return 0;
2022 
2023 	thr = 0;
2024 	while (++thr < threads_per_subcore)
2025 		if (cpu_online(cpu + thr))
2026 			return 0;
2027 
2028 	/* Grab all hw threads so they can't go into the kernel */
2029 	for (thr = 1; thr < threads_per_subcore; ++thr) {
2030 		if (kvmppc_grab_hwthread(cpu + thr)) {
2031 			/* Couldn't grab one; let the others go */
2032 			do {
2033 				kvmppc_release_hwthread(cpu + thr);
2034 			} while (--thr > 0);
2035 			return 0;
2036 		}
2037 	}
2038 	return 1;
2039 }
2040 
2041 /*
2042  * A list of virtual cores for each physical CPU.
2043  * These are vcores that could run but their runner VCPU tasks are
2044  * (or may be) preempted.
2045  */
2046 struct preempted_vcore_list {
2047 	struct list_head	list;
2048 	spinlock_t		lock;
2049 };
2050 
2051 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2052 
2053 static void init_vcore_lists(void)
2054 {
2055 	int cpu;
2056 
2057 	for_each_possible_cpu(cpu) {
2058 		struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2059 		spin_lock_init(&lp->lock);
2060 		INIT_LIST_HEAD(&lp->list);
2061 	}
2062 }
2063 
2064 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2065 {
2066 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2067 
2068 	vc->vcore_state = VCORE_PREEMPT;
2069 	vc->pcpu = smp_processor_id();
2070 	if (vc->num_threads < threads_per_vcore()) {
2071 		spin_lock(&lp->lock);
2072 		list_add_tail(&vc->preempt_list, &lp->list);
2073 		spin_unlock(&lp->lock);
2074 	}
2075 
2076 	/* Start accumulating stolen time */
2077 	kvmppc_core_start_stolen(vc);
2078 }
2079 
2080 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2081 {
2082 	struct preempted_vcore_list *lp;
2083 
2084 	kvmppc_core_end_stolen(vc);
2085 	if (!list_empty(&vc->preempt_list)) {
2086 		lp = &per_cpu(preempted_vcores, vc->pcpu);
2087 		spin_lock(&lp->lock);
2088 		list_del_init(&vc->preempt_list);
2089 		spin_unlock(&lp->lock);
2090 	}
2091 	vc->vcore_state = VCORE_INACTIVE;
2092 }
2093 
2094 /*
2095  * This stores information about the virtual cores currently
2096  * assigned to a physical core.
2097  */
2098 struct core_info {
2099 	int		n_subcores;
2100 	int		max_subcore_threads;
2101 	int		total_threads;
2102 	int		subcore_threads[MAX_SUBCORES];
2103 	struct kvm	*subcore_vm[MAX_SUBCORES];
2104 	struct list_head vcs[MAX_SUBCORES];
2105 };
2106 
2107 /*
2108  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2109  * respectively in 2-way micro-threading (split-core) mode.
2110  */
2111 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2112 
2113 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2114 {
2115 	int sub;
2116 
2117 	memset(cip, 0, sizeof(*cip));
2118 	cip->n_subcores = 1;
2119 	cip->max_subcore_threads = vc->num_threads;
2120 	cip->total_threads = vc->num_threads;
2121 	cip->subcore_threads[0] = vc->num_threads;
2122 	cip->subcore_vm[0] = vc->kvm;
2123 	for (sub = 0; sub < MAX_SUBCORES; ++sub)
2124 		INIT_LIST_HEAD(&cip->vcs[sub]);
2125 	list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2126 }
2127 
2128 static bool subcore_config_ok(int n_subcores, int n_threads)
2129 {
2130 	/* Can only dynamically split if unsplit to begin with */
2131 	if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2132 		return false;
2133 	if (n_subcores > MAX_SUBCORES)
2134 		return false;
2135 	if (n_subcores > 1) {
2136 		if (!(dynamic_mt_modes & 2))
2137 			n_subcores = 4;
2138 		if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2139 			return false;
2140 	}
2141 
2142 	return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2143 }
2144 
2145 static void init_master_vcore(struct kvmppc_vcore *vc)
2146 {
2147 	vc->master_vcore = vc;
2148 	vc->entry_exit_map = 0;
2149 	vc->in_guest = 0;
2150 	vc->napping_threads = 0;
2151 	vc->conferring_threads = 0;
2152 }
2153 
2154 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2155 {
2156 	int n_threads = vc->num_threads;
2157 	int sub;
2158 
2159 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2160 		return false;
2161 
2162 	if (n_threads < cip->max_subcore_threads)
2163 		n_threads = cip->max_subcore_threads;
2164 	if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2165 		return false;
2166 	cip->max_subcore_threads = n_threads;
2167 
2168 	sub = cip->n_subcores;
2169 	++cip->n_subcores;
2170 	cip->total_threads += vc->num_threads;
2171 	cip->subcore_threads[sub] = vc->num_threads;
2172 	cip->subcore_vm[sub] = vc->kvm;
2173 	init_master_vcore(vc);
2174 	list_move_tail(&vc->preempt_list, &cip->vcs[sub]);
2175 
2176 	return true;
2177 }
2178 
2179 /*
2180  * Work out whether it is possible to piggyback the execution of
2181  * vcore *pvc onto the execution of the other vcores described in *cip.
2182  */
2183 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2184 			  int target_threads)
2185 {
2186 	if (cip->total_threads + pvc->num_threads > target_threads)
2187 		return false;
2188 
2189 	return can_dynamic_split(pvc, cip);
2190 }
2191 
2192 static void prepare_threads(struct kvmppc_vcore *vc)
2193 {
2194 	int i;
2195 	struct kvm_vcpu *vcpu;
2196 
2197 	for_each_runnable_thread(i, vcpu, vc) {
2198 		if (signal_pending(vcpu->arch.run_task))
2199 			vcpu->arch.ret = -EINTR;
2200 		else if (vcpu->arch.vpa.update_pending ||
2201 			 vcpu->arch.slb_shadow.update_pending ||
2202 			 vcpu->arch.dtl.update_pending)
2203 			vcpu->arch.ret = RESUME_GUEST;
2204 		else
2205 			continue;
2206 		kvmppc_remove_runnable(vc, vcpu);
2207 		wake_up(&vcpu->arch.cpu_run);
2208 	}
2209 }
2210 
2211 static void collect_piggybacks(struct core_info *cip, int target_threads)
2212 {
2213 	struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2214 	struct kvmppc_vcore *pvc, *vcnext;
2215 
2216 	spin_lock(&lp->lock);
2217 	list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2218 		if (!spin_trylock(&pvc->lock))
2219 			continue;
2220 		prepare_threads(pvc);
2221 		if (!pvc->n_runnable) {
2222 			list_del_init(&pvc->preempt_list);
2223 			if (pvc->runner == NULL) {
2224 				pvc->vcore_state = VCORE_INACTIVE;
2225 				kvmppc_core_end_stolen(pvc);
2226 			}
2227 			spin_unlock(&pvc->lock);
2228 			continue;
2229 		}
2230 		if (!can_piggyback(pvc, cip, target_threads)) {
2231 			spin_unlock(&pvc->lock);
2232 			continue;
2233 		}
2234 		kvmppc_core_end_stolen(pvc);
2235 		pvc->vcore_state = VCORE_PIGGYBACK;
2236 		if (cip->total_threads >= target_threads)
2237 			break;
2238 	}
2239 	spin_unlock(&lp->lock);
2240 }
2241 
2242 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2243 {
2244 	int still_running = 0, i;
2245 	u64 now;
2246 	long ret;
2247 	struct kvm_vcpu *vcpu;
2248 
2249 	spin_lock(&vc->lock);
2250 	now = get_tb();
2251 	for_each_runnable_thread(i, vcpu, vc) {
2252 		/* cancel pending dec exception if dec is positive */
2253 		if (now < vcpu->arch.dec_expires &&
2254 		    kvmppc_core_pending_dec(vcpu))
2255 			kvmppc_core_dequeue_dec(vcpu);
2256 
2257 		trace_kvm_guest_exit(vcpu);
2258 
2259 		ret = RESUME_GUEST;
2260 		if (vcpu->arch.trap)
2261 			ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2262 						    vcpu->arch.run_task);
2263 
2264 		vcpu->arch.ret = ret;
2265 		vcpu->arch.trap = 0;
2266 
2267 		if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2268 			if (vcpu->arch.pending_exceptions)
2269 				kvmppc_core_prepare_to_enter(vcpu);
2270 			if (vcpu->arch.ceded)
2271 				kvmppc_set_timer(vcpu);
2272 			else
2273 				++still_running;
2274 		} else {
2275 			kvmppc_remove_runnable(vc, vcpu);
2276 			wake_up(&vcpu->arch.cpu_run);
2277 		}
2278 	}
2279 	list_del_init(&vc->preempt_list);
2280 	if (!is_master) {
2281 		if (still_running > 0) {
2282 			kvmppc_vcore_preempt(vc);
2283 		} else if (vc->runner) {
2284 			vc->vcore_state = VCORE_PREEMPT;
2285 			kvmppc_core_start_stolen(vc);
2286 		} else {
2287 			vc->vcore_state = VCORE_INACTIVE;
2288 		}
2289 		if (vc->n_runnable > 0 && vc->runner == NULL) {
2290 			/* make sure there's a candidate runner awake */
2291 			i = -1;
2292 			vcpu = next_runnable_thread(vc, &i);
2293 			wake_up(&vcpu->arch.cpu_run);
2294 		}
2295 	}
2296 	spin_unlock(&vc->lock);
2297 }
2298 
2299 /*
2300  * Clear core from the list of active host cores as we are about to
2301  * enter the guest. Only do this if it is the primary thread of the
2302  * core (not if a subcore) that is entering the guest.
2303  */
2304 static inline int kvmppc_clear_host_core(unsigned int cpu)
2305 {
2306 	int core;
2307 
2308 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2309 		return 0;
2310 	/*
2311 	 * Memory barrier can be omitted here as we will do a smp_wmb()
2312 	 * later in kvmppc_start_thread and we need ensure that state is
2313 	 * visible to other CPUs only after we enter guest.
2314 	 */
2315 	core = cpu >> threads_shift;
2316 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2317 	return 0;
2318 }
2319 
2320 /*
2321  * Advertise this core as an active host core since we exited the guest
2322  * Only need to do this if it is the primary thread of the core that is
2323  * exiting.
2324  */
2325 static inline int kvmppc_set_host_core(unsigned int cpu)
2326 {
2327 	int core;
2328 
2329 	if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2330 		return 0;
2331 
2332 	/*
2333 	 * Memory barrier can be omitted here because we do a spin_unlock
2334 	 * immediately after this which provides the memory barrier.
2335 	 */
2336 	core = cpu >> threads_shift;
2337 	kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2338 	return 0;
2339 }
2340 
2341 /*
2342  * Run a set of guest threads on a physical core.
2343  * Called with vc->lock held.
2344  */
2345 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2346 {
2347 	struct kvm_vcpu *vcpu;
2348 	int i;
2349 	int srcu_idx;
2350 	struct core_info core_info;
2351 	struct kvmppc_vcore *pvc, *vcnext;
2352 	struct kvm_split_mode split_info, *sip;
2353 	int split, subcore_size, active;
2354 	int sub;
2355 	bool thr0_done;
2356 	unsigned long cmd_bit, stat_bit;
2357 	int pcpu, thr;
2358 	int target_threads;
2359 	int controlled_threads;
2360 
2361 	/*
2362 	 * Remove from the list any threads that have a signal pending
2363 	 * or need a VPA update done
2364 	 */
2365 	prepare_threads(vc);
2366 
2367 	/* if the runner is no longer runnable, let the caller pick a new one */
2368 	if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2369 		return;
2370 
2371 	/*
2372 	 * Initialize *vc.
2373 	 */
2374 	init_master_vcore(vc);
2375 	vc->preempt_tb = TB_NIL;
2376 
2377 	/*
2378 	 * Number of threads that we will be controlling: the same as
2379 	 * the number of threads per subcore, except on POWER9,
2380 	 * where it's 1 because the threads are (mostly) independent.
2381 	 */
2382 	controlled_threads = threads_per_vcore();
2383 
2384 	/*
2385 	 * Make sure we are running on primary threads, and that secondary
2386 	 * threads are offline.  Also check if the number of threads in this
2387 	 * guest are greater than the current system threads per guest.
2388 	 */
2389 	if ((controlled_threads > 1) &&
2390 	    ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2391 		for_each_runnable_thread(i, vcpu, vc) {
2392 			vcpu->arch.ret = -EBUSY;
2393 			kvmppc_remove_runnable(vc, vcpu);
2394 			wake_up(&vcpu->arch.cpu_run);
2395 		}
2396 		goto out;
2397 	}
2398 
2399 	/*
2400 	 * See if we could run any other vcores on the physical core
2401 	 * along with this one.
2402 	 */
2403 	init_core_info(&core_info, vc);
2404 	pcpu = smp_processor_id();
2405 	target_threads = controlled_threads;
2406 	if (target_smt_mode && target_smt_mode < target_threads)
2407 		target_threads = target_smt_mode;
2408 	if (vc->num_threads < target_threads)
2409 		collect_piggybacks(&core_info, target_threads);
2410 
2411 	/* Decide on micro-threading (split-core) mode */
2412 	subcore_size = threads_per_subcore;
2413 	cmd_bit = stat_bit = 0;
2414 	split = core_info.n_subcores;
2415 	sip = NULL;
2416 	if (split > 1) {
2417 		/* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2418 		if (split == 2 && (dynamic_mt_modes & 2)) {
2419 			cmd_bit = HID0_POWER8_1TO2LPAR;
2420 			stat_bit = HID0_POWER8_2LPARMODE;
2421 		} else {
2422 			split = 4;
2423 			cmd_bit = HID0_POWER8_1TO4LPAR;
2424 			stat_bit = HID0_POWER8_4LPARMODE;
2425 		}
2426 		subcore_size = MAX_SMT_THREADS / split;
2427 		sip = &split_info;
2428 		memset(&split_info, 0, sizeof(split_info));
2429 		split_info.rpr = mfspr(SPRN_RPR);
2430 		split_info.pmmar = mfspr(SPRN_PMMAR);
2431 		split_info.ldbar = mfspr(SPRN_LDBAR);
2432 		split_info.subcore_size = subcore_size;
2433 		for (sub = 0; sub < core_info.n_subcores; ++sub)
2434 			split_info.master_vcs[sub] =
2435 				list_first_entry(&core_info.vcs[sub],
2436 					struct kvmppc_vcore, preempt_list);
2437 		/* order writes to split_info before kvm_split_mode pointer */
2438 		smp_wmb();
2439 	}
2440 	pcpu = smp_processor_id();
2441 	for (thr = 0; thr < controlled_threads; ++thr)
2442 		paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2443 
2444 	/* Initiate micro-threading (split-core) if required */
2445 	if (cmd_bit) {
2446 		unsigned long hid0 = mfspr(SPRN_HID0);
2447 
2448 		hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2449 		mb();
2450 		mtspr(SPRN_HID0, hid0);
2451 		isync();
2452 		for (;;) {
2453 			hid0 = mfspr(SPRN_HID0);
2454 			if (hid0 & stat_bit)
2455 				break;
2456 			cpu_relax();
2457 		}
2458 	}
2459 
2460 	kvmppc_clear_host_core(pcpu);
2461 
2462 	/* Start all the threads */
2463 	active = 0;
2464 	for (sub = 0; sub < core_info.n_subcores; ++sub) {
2465 		thr = subcore_thread_map[sub];
2466 		thr0_done = false;
2467 		active |= 1 << thr;
2468 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2469 			pvc->pcpu = pcpu + thr;
2470 			for_each_runnable_thread(i, vcpu, pvc) {
2471 				kvmppc_start_thread(vcpu, pvc);
2472 				kvmppc_create_dtl_entry(vcpu, pvc);
2473 				trace_kvm_guest_enter(vcpu);
2474 				if (!vcpu->arch.ptid)
2475 					thr0_done = true;
2476 				active |= 1 << (thr + vcpu->arch.ptid);
2477 			}
2478 			/*
2479 			 * We need to start the first thread of each subcore
2480 			 * even if it doesn't have a vcpu.
2481 			 */
2482 			if (pvc->master_vcore == pvc && !thr0_done)
2483 				kvmppc_start_thread(NULL, pvc);
2484 			thr += pvc->num_threads;
2485 		}
2486 	}
2487 
2488 	/*
2489 	 * Ensure that split_info.do_nap is set after setting
2490 	 * the vcore pointer in the PACA of the secondaries.
2491 	 */
2492 	smp_mb();
2493 	if (cmd_bit)
2494 		split_info.do_nap = 1;	/* ask secondaries to nap when done */
2495 
2496 	/*
2497 	 * When doing micro-threading, poke the inactive threads as well.
2498 	 * This gets them to the nap instruction after kvm_do_nap,
2499 	 * which reduces the time taken to unsplit later.
2500 	 */
2501 	if (split > 1)
2502 		for (thr = 1; thr < threads_per_subcore; ++thr)
2503 			if (!(active & (1 << thr)))
2504 				kvmppc_ipi_thread(pcpu + thr);
2505 
2506 	vc->vcore_state = VCORE_RUNNING;
2507 	preempt_disable();
2508 
2509 	trace_kvmppc_run_core(vc, 0);
2510 
2511 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2512 		list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2513 			spin_unlock(&pvc->lock);
2514 
2515 	guest_enter();
2516 
2517 	srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2518 
2519 	__kvmppc_vcore_entry();
2520 
2521 	srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2522 
2523 	spin_lock(&vc->lock);
2524 	/* prevent other vcpu threads from doing kvmppc_start_thread() now */
2525 	vc->vcore_state = VCORE_EXITING;
2526 
2527 	/* wait for secondary threads to finish writing their state to memory */
2528 	kvmppc_wait_for_nap();
2529 
2530 	/* Return to whole-core mode if we split the core earlier */
2531 	if (split > 1) {
2532 		unsigned long hid0 = mfspr(SPRN_HID0);
2533 		unsigned long loops = 0;
2534 
2535 		hid0 &= ~HID0_POWER8_DYNLPARDIS;
2536 		stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2537 		mb();
2538 		mtspr(SPRN_HID0, hid0);
2539 		isync();
2540 		for (;;) {
2541 			hid0 = mfspr(SPRN_HID0);
2542 			if (!(hid0 & stat_bit))
2543 				break;
2544 			cpu_relax();
2545 			++loops;
2546 		}
2547 		split_info.do_nap = 0;
2548 	}
2549 
2550 	/* Let secondaries go back to the offline loop */
2551 	for (i = 0; i < controlled_threads; ++i) {
2552 		kvmppc_release_hwthread(pcpu + i);
2553 		if (sip && sip->napped[i])
2554 			kvmppc_ipi_thread(pcpu + i);
2555 	}
2556 
2557 	kvmppc_set_host_core(pcpu);
2558 
2559 	spin_unlock(&vc->lock);
2560 
2561 	/* make sure updates to secondary vcpu structs are visible now */
2562 	smp_mb();
2563 	guest_exit();
2564 
2565 	for (sub = 0; sub < core_info.n_subcores; ++sub)
2566 		list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2567 					 preempt_list)
2568 			post_guest_process(pvc, pvc == vc);
2569 
2570 	spin_lock(&vc->lock);
2571 	preempt_enable();
2572 
2573  out:
2574 	vc->vcore_state = VCORE_INACTIVE;
2575 	trace_kvmppc_run_core(vc, 1);
2576 }
2577 
2578 /*
2579  * Wait for some other vcpu thread to execute us, and
2580  * wake us up when we need to handle something in the host.
2581  */
2582 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2583 				 struct kvm_vcpu *vcpu, int wait_state)
2584 {
2585 	DEFINE_WAIT(wait);
2586 
2587 	prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2588 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2589 		spin_unlock(&vc->lock);
2590 		schedule();
2591 		spin_lock(&vc->lock);
2592 	}
2593 	finish_wait(&vcpu->arch.cpu_run, &wait);
2594 }
2595 
2596 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2597 {
2598 	/* 10us base */
2599 	if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2600 		vc->halt_poll_ns = 10000;
2601 	else
2602 		vc->halt_poll_ns *= halt_poll_ns_grow;
2603 }
2604 
2605 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2606 {
2607 	if (halt_poll_ns_shrink == 0)
2608 		vc->halt_poll_ns = 0;
2609 	else
2610 		vc->halt_poll_ns /= halt_poll_ns_shrink;
2611 }
2612 
2613 /*
2614  * Check to see if any of the runnable vcpus on the vcore have pending
2615  * exceptions or are no longer ceded
2616  */
2617 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2618 {
2619 	struct kvm_vcpu *vcpu;
2620 	int i;
2621 
2622 	for_each_runnable_thread(i, vcpu, vc) {
2623 		if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded)
2624 			return 1;
2625 	}
2626 
2627 	return 0;
2628 }
2629 
2630 /*
2631  * All the vcpus in this vcore are idle, so wait for a decrementer
2632  * or external interrupt to one of the vcpus.  vc->lock is held.
2633  */
2634 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2635 {
2636 	ktime_t cur, start_poll, start_wait;
2637 	int do_sleep = 1;
2638 	u64 block_ns;
2639 	DECLARE_SWAITQUEUE(wait);
2640 
2641 	/* Poll for pending exceptions and ceded state */
2642 	cur = start_poll = ktime_get();
2643 	if (vc->halt_poll_ns) {
2644 		ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2645 		++vc->runner->stat.halt_attempted_poll;
2646 
2647 		vc->vcore_state = VCORE_POLLING;
2648 		spin_unlock(&vc->lock);
2649 
2650 		do {
2651 			if (kvmppc_vcore_check_block(vc)) {
2652 				do_sleep = 0;
2653 				break;
2654 			}
2655 			cur = ktime_get();
2656 		} while (single_task_running() && ktime_before(cur, stop));
2657 
2658 		spin_lock(&vc->lock);
2659 		vc->vcore_state = VCORE_INACTIVE;
2660 
2661 		if (!do_sleep) {
2662 			++vc->runner->stat.halt_successful_poll;
2663 			goto out;
2664 		}
2665 	}
2666 
2667 	prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2668 
2669 	if (kvmppc_vcore_check_block(vc)) {
2670 		finish_swait(&vc->wq, &wait);
2671 		do_sleep = 0;
2672 		/* If we polled, count this as a successful poll */
2673 		if (vc->halt_poll_ns)
2674 			++vc->runner->stat.halt_successful_poll;
2675 		goto out;
2676 	}
2677 
2678 	start_wait = ktime_get();
2679 
2680 	vc->vcore_state = VCORE_SLEEPING;
2681 	trace_kvmppc_vcore_blocked(vc, 0);
2682 	spin_unlock(&vc->lock);
2683 	schedule();
2684 	finish_swait(&vc->wq, &wait);
2685 	spin_lock(&vc->lock);
2686 	vc->vcore_state = VCORE_INACTIVE;
2687 	trace_kvmppc_vcore_blocked(vc, 1);
2688 	++vc->runner->stat.halt_successful_wait;
2689 
2690 	cur = ktime_get();
2691 
2692 out:
2693 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2694 
2695 	/* Attribute wait time */
2696 	if (do_sleep) {
2697 		vc->runner->stat.halt_wait_ns +=
2698 			ktime_to_ns(cur) - ktime_to_ns(start_wait);
2699 		/* Attribute failed poll time */
2700 		if (vc->halt_poll_ns)
2701 			vc->runner->stat.halt_poll_fail_ns +=
2702 				ktime_to_ns(start_wait) -
2703 				ktime_to_ns(start_poll);
2704 	} else {
2705 		/* Attribute successful poll time */
2706 		if (vc->halt_poll_ns)
2707 			vc->runner->stat.halt_poll_success_ns +=
2708 				ktime_to_ns(cur) -
2709 				ktime_to_ns(start_poll);
2710 	}
2711 
2712 	/* Adjust poll time */
2713 	if (halt_poll_ns) {
2714 		if (block_ns <= vc->halt_poll_ns)
2715 			;
2716 		/* We slept and blocked for longer than the max halt time */
2717 		else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
2718 			shrink_halt_poll_ns(vc);
2719 		/* We slept and our poll time is too small */
2720 		else if (vc->halt_poll_ns < halt_poll_ns &&
2721 				block_ns < halt_poll_ns)
2722 			grow_halt_poll_ns(vc);
2723 		if (vc->halt_poll_ns > halt_poll_ns)
2724 			vc->halt_poll_ns = halt_poll_ns;
2725 	} else
2726 		vc->halt_poll_ns = 0;
2727 
2728 	trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2729 }
2730 
2731 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2732 {
2733 	int n_ceded, i;
2734 	struct kvmppc_vcore *vc;
2735 	struct kvm_vcpu *v;
2736 
2737 	trace_kvmppc_run_vcpu_enter(vcpu);
2738 
2739 	kvm_run->exit_reason = 0;
2740 	vcpu->arch.ret = RESUME_GUEST;
2741 	vcpu->arch.trap = 0;
2742 	kvmppc_update_vpas(vcpu);
2743 
2744 	/*
2745 	 * Synchronize with other threads in this virtual core
2746 	 */
2747 	vc = vcpu->arch.vcore;
2748 	spin_lock(&vc->lock);
2749 	vcpu->arch.ceded = 0;
2750 	vcpu->arch.run_task = current;
2751 	vcpu->arch.kvm_run = kvm_run;
2752 	vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2753 	vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2754 	vcpu->arch.busy_preempt = TB_NIL;
2755 	WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2756 	++vc->n_runnable;
2757 
2758 	/*
2759 	 * This happens the first time this is called for a vcpu.
2760 	 * If the vcore is already running, we may be able to start
2761 	 * this thread straight away and have it join in.
2762 	 */
2763 	if (!signal_pending(current)) {
2764 		if (vc->vcore_state == VCORE_PIGGYBACK) {
2765 			struct kvmppc_vcore *mvc = vc->master_vcore;
2766 			if (spin_trylock(&mvc->lock)) {
2767 				if (mvc->vcore_state == VCORE_RUNNING &&
2768 				    !VCORE_IS_EXITING(mvc)) {
2769 					kvmppc_create_dtl_entry(vcpu, vc);
2770 					kvmppc_start_thread(vcpu, vc);
2771 					trace_kvm_guest_enter(vcpu);
2772 				}
2773 				spin_unlock(&mvc->lock);
2774 			}
2775 		} else if (vc->vcore_state == VCORE_RUNNING &&
2776 			   !VCORE_IS_EXITING(vc)) {
2777 			kvmppc_create_dtl_entry(vcpu, vc);
2778 			kvmppc_start_thread(vcpu, vc);
2779 			trace_kvm_guest_enter(vcpu);
2780 		} else if (vc->vcore_state == VCORE_SLEEPING) {
2781 			swake_up(&vc->wq);
2782 		}
2783 
2784 	}
2785 
2786 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2787 	       !signal_pending(current)) {
2788 		if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2789 			kvmppc_vcore_end_preempt(vc);
2790 
2791 		if (vc->vcore_state != VCORE_INACTIVE) {
2792 			kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2793 			continue;
2794 		}
2795 		for_each_runnable_thread(i, v, vc) {
2796 			kvmppc_core_prepare_to_enter(v);
2797 			if (signal_pending(v->arch.run_task)) {
2798 				kvmppc_remove_runnable(vc, v);
2799 				v->stat.signal_exits++;
2800 				v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2801 				v->arch.ret = -EINTR;
2802 				wake_up(&v->arch.cpu_run);
2803 			}
2804 		}
2805 		if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2806 			break;
2807 		n_ceded = 0;
2808 		for_each_runnable_thread(i, v, vc) {
2809 			if (!v->arch.pending_exceptions)
2810 				n_ceded += v->arch.ceded;
2811 			else
2812 				v->arch.ceded = 0;
2813 		}
2814 		vc->runner = vcpu;
2815 		if (n_ceded == vc->n_runnable) {
2816 			kvmppc_vcore_blocked(vc);
2817 		} else if (need_resched()) {
2818 			kvmppc_vcore_preempt(vc);
2819 			/* Let something else run */
2820 			cond_resched_lock(&vc->lock);
2821 			if (vc->vcore_state == VCORE_PREEMPT)
2822 				kvmppc_vcore_end_preempt(vc);
2823 		} else {
2824 			kvmppc_run_core(vc);
2825 		}
2826 		vc->runner = NULL;
2827 	}
2828 
2829 	while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2830 	       (vc->vcore_state == VCORE_RUNNING ||
2831 		vc->vcore_state == VCORE_EXITING ||
2832 		vc->vcore_state == VCORE_PIGGYBACK))
2833 		kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2834 
2835 	if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2836 		kvmppc_vcore_end_preempt(vc);
2837 
2838 	if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2839 		kvmppc_remove_runnable(vc, vcpu);
2840 		vcpu->stat.signal_exits++;
2841 		kvm_run->exit_reason = KVM_EXIT_INTR;
2842 		vcpu->arch.ret = -EINTR;
2843 	}
2844 
2845 	if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2846 		/* Wake up some vcpu to run the core */
2847 		i = -1;
2848 		v = next_runnable_thread(vc, &i);
2849 		wake_up(&v->arch.cpu_run);
2850 	}
2851 
2852 	trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2853 	spin_unlock(&vc->lock);
2854 	return vcpu->arch.ret;
2855 }
2856 
2857 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2858 {
2859 	int r;
2860 	int srcu_idx;
2861 
2862 	if (!vcpu->arch.sane) {
2863 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2864 		return -EINVAL;
2865 	}
2866 
2867 	kvmppc_core_prepare_to_enter(vcpu);
2868 
2869 	/* No need to go into the guest when all we'll do is come back out */
2870 	if (signal_pending(current)) {
2871 		run->exit_reason = KVM_EXIT_INTR;
2872 		return -EINTR;
2873 	}
2874 
2875 	atomic_inc(&vcpu->kvm->arch.vcpus_running);
2876 	/* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2877 	smp_mb();
2878 
2879 	/* On the first time here, set up HTAB and VRMA */
2880 	if (!vcpu->kvm->arch.hpte_setup_done) {
2881 		r = kvmppc_hv_setup_htab_rma(vcpu);
2882 		if (r)
2883 			goto out;
2884 	}
2885 
2886 	flush_all_to_thread(current);
2887 
2888 	vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2889 	vcpu->arch.pgdir = current->mm->pgd;
2890 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2891 
2892 	do {
2893 		r = kvmppc_run_vcpu(run, vcpu);
2894 
2895 		if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2896 		    !(vcpu->arch.shregs.msr & MSR_PR)) {
2897 			trace_kvm_hcall_enter(vcpu);
2898 			r = kvmppc_pseries_do_hcall(vcpu);
2899 			trace_kvm_hcall_exit(vcpu, r);
2900 			kvmppc_core_prepare_to_enter(vcpu);
2901 		} else if (r == RESUME_PAGE_FAULT) {
2902 			srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2903 			r = kvmppc_book3s_hv_page_fault(run, vcpu,
2904 				vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2905 			srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2906 		} else if (r == RESUME_PASSTHROUGH)
2907 			r = kvmppc_xics_rm_complete(vcpu, 0);
2908 	} while (is_kvmppc_resume_guest(r));
2909 
2910  out:
2911 	vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2912 	atomic_dec(&vcpu->kvm->arch.vcpus_running);
2913 	return r;
2914 }
2915 
2916 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2917 				     int linux_psize)
2918 {
2919 	struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2920 
2921 	if (!def->shift)
2922 		return;
2923 	(*sps)->page_shift = def->shift;
2924 	(*sps)->slb_enc = def->sllp;
2925 	(*sps)->enc[0].page_shift = def->shift;
2926 	(*sps)->enc[0].pte_enc = def->penc[linux_psize];
2927 	/*
2928 	 * Add 16MB MPSS support if host supports it
2929 	 */
2930 	if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2931 		(*sps)->enc[1].page_shift = 24;
2932 		(*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2933 	}
2934 	(*sps)++;
2935 }
2936 
2937 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2938 					 struct kvm_ppc_smmu_info *info)
2939 {
2940 	struct kvm_ppc_one_seg_page_size *sps;
2941 
2942 	info->flags = KVM_PPC_PAGE_SIZES_REAL;
2943 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2944 		info->flags |= KVM_PPC_1T_SEGMENTS;
2945 	info->slb_size = mmu_slb_size;
2946 
2947 	/* We only support these sizes for now, and no muti-size segments */
2948 	sps = &info->sps[0];
2949 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
2950 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
2951 	kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
2952 
2953 	return 0;
2954 }
2955 
2956 /*
2957  * Get (and clear) the dirty memory log for a memory slot.
2958  */
2959 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
2960 					 struct kvm_dirty_log *log)
2961 {
2962 	struct kvm_memslots *slots;
2963 	struct kvm_memory_slot *memslot;
2964 	int r;
2965 	unsigned long n;
2966 
2967 	mutex_lock(&kvm->slots_lock);
2968 
2969 	r = -EINVAL;
2970 	if (log->slot >= KVM_USER_MEM_SLOTS)
2971 		goto out;
2972 
2973 	slots = kvm_memslots(kvm);
2974 	memslot = id_to_memslot(slots, log->slot);
2975 	r = -ENOENT;
2976 	if (!memslot->dirty_bitmap)
2977 		goto out;
2978 
2979 	n = kvm_dirty_bitmap_bytes(memslot);
2980 	memset(memslot->dirty_bitmap, 0, n);
2981 
2982 	r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap);
2983 	if (r)
2984 		goto out;
2985 
2986 	r = -EFAULT;
2987 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
2988 		goto out;
2989 
2990 	r = 0;
2991 out:
2992 	mutex_unlock(&kvm->slots_lock);
2993 	return r;
2994 }
2995 
2996 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
2997 					struct kvm_memory_slot *dont)
2998 {
2999 	if (!dont || free->arch.rmap != dont->arch.rmap) {
3000 		vfree(free->arch.rmap);
3001 		free->arch.rmap = NULL;
3002 	}
3003 }
3004 
3005 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3006 					 unsigned long npages)
3007 {
3008 	slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3009 	if (!slot->arch.rmap)
3010 		return -ENOMEM;
3011 
3012 	return 0;
3013 }
3014 
3015 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3016 					struct kvm_memory_slot *memslot,
3017 					const struct kvm_userspace_memory_region *mem)
3018 {
3019 	return 0;
3020 }
3021 
3022 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3023 				const struct kvm_userspace_memory_region *mem,
3024 				const struct kvm_memory_slot *old,
3025 				const struct kvm_memory_slot *new)
3026 {
3027 	unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3028 	struct kvm_memslots *slots;
3029 	struct kvm_memory_slot *memslot;
3030 
3031 	/*
3032 	 * If we are making a new memslot, it might make
3033 	 * some address that was previously cached as emulated
3034 	 * MMIO be no longer emulated MMIO, so invalidate
3035 	 * all the caches of emulated MMIO translations.
3036 	 */
3037 	if (npages)
3038 		atomic64_inc(&kvm->arch.mmio_update);
3039 
3040 	if (npages && old->npages) {
3041 		/*
3042 		 * If modifying a memslot, reset all the rmap dirty bits.
3043 		 * If this is a new memslot, we don't need to do anything
3044 		 * since the rmap array starts out as all zeroes,
3045 		 * i.e. no pages are dirty.
3046 		 */
3047 		slots = kvm_memslots(kvm);
3048 		memslot = id_to_memslot(slots, mem->slot);
3049 		kvmppc_hv_get_dirty_log(kvm, memslot, NULL);
3050 	}
3051 }
3052 
3053 /*
3054  * Update LPCR values in kvm->arch and in vcores.
3055  * Caller must hold kvm->lock.
3056  */
3057 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3058 {
3059 	long int i;
3060 	u32 cores_done = 0;
3061 
3062 	if ((kvm->arch.lpcr & mask) == lpcr)
3063 		return;
3064 
3065 	kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3066 
3067 	for (i = 0; i < KVM_MAX_VCORES; ++i) {
3068 		struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3069 		if (!vc)
3070 			continue;
3071 		spin_lock(&vc->lock);
3072 		vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3073 		spin_unlock(&vc->lock);
3074 		if (++cores_done >= kvm->arch.online_vcores)
3075 			break;
3076 	}
3077 }
3078 
3079 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3080 {
3081 	return;
3082 }
3083 
3084 static void kvmppc_setup_partition_table(struct kvm *kvm)
3085 {
3086 	unsigned long dw0, dw1;
3087 
3088 	/* PS field - page size for VRMA */
3089 	dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3090 		((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3091 	/* HTABSIZE and HTABORG fields */
3092 	dw0 |= kvm->arch.sdr1;
3093 
3094 	/* Second dword has GR=0; other fields are unused since UPRT=0 */
3095 	dw1 = 0;
3096 
3097 	mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3098 }
3099 
3100 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3101 {
3102 	int err = 0;
3103 	struct kvm *kvm = vcpu->kvm;
3104 	unsigned long hva;
3105 	struct kvm_memory_slot *memslot;
3106 	struct vm_area_struct *vma;
3107 	unsigned long lpcr = 0, senc;
3108 	unsigned long psize, porder;
3109 	int srcu_idx;
3110 
3111 	mutex_lock(&kvm->lock);
3112 	if (kvm->arch.hpte_setup_done)
3113 		goto out;	/* another vcpu beat us to it */
3114 
3115 	/* Allocate hashed page table (if not done already) and reset it */
3116 	if (!kvm->arch.hpt_virt) {
3117 		err = kvmppc_alloc_hpt(kvm, NULL);
3118 		if (err) {
3119 			pr_err("KVM: Couldn't alloc HPT\n");
3120 			goto out;
3121 		}
3122 	}
3123 
3124 	/* Look up the memslot for guest physical address 0 */
3125 	srcu_idx = srcu_read_lock(&kvm->srcu);
3126 	memslot = gfn_to_memslot(kvm, 0);
3127 
3128 	/* We must have some memory at 0 by now */
3129 	err = -EINVAL;
3130 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3131 		goto out_srcu;
3132 
3133 	/* Look up the VMA for the start of this memory slot */
3134 	hva = memslot->userspace_addr;
3135 	down_read(&current->mm->mmap_sem);
3136 	vma = find_vma(current->mm, hva);
3137 	if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3138 		goto up_out;
3139 
3140 	psize = vma_kernel_pagesize(vma);
3141 	porder = __ilog2(psize);
3142 
3143 	up_read(&current->mm->mmap_sem);
3144 
3145 	/* We can handle 4k, 64k or 16M pages in the VRMA */
3146 	err = -EINVAL;
3147 	if (!(psize == 0x1000 || psize == 0x10000 ||
3148 	      psize == 0x1000000))
3149 		goto out_srcu;
3150 
3151 	senc = slb_pgsize_encoding(psize);
3152 	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3153 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3154 	/* Create HPTEs in the hash page table for the VRMA */
3155 	kvmppc_map_vrma(vcpu, memslot, porder);
3156 
3157 	/* Update VRMASD field in the LPCR */
3158 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3159 		/* the -4 is to account for senc values starting at 0x10 */
3160 		lpcr = senc << (LPCR_VRMASD_SH - 4);
3161 		kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3162 	} else {
3163 		kvmppc_setup_partition_table(kvm);
3164 	}
3165 
3166 	/* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3167 	smp_wmb();
3168 	kvm->arch.hpte_setup_done = 1;
3169 	err = 0;
3170  out_srcu:
3171 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3172  out:
3173 	mutex_unlock(&kvm->lock);
3174 	return err;
3175 
3176  up_out:
3177 	up_read(&current->mm->mmap_sem);
3178 	goto out_srcu;
3179 }
3180 
3181 #ifdef CONFIG_KVM_XICS
3182 /*
3183  * Allocate a per-core structure for managing state about which cores are
3184  * running in the host versus the guest and for exchanging data between
3185  * real mode KVM and CPU running in the host.
3186  * This is only done for the first VM.
3187  * The allocated structure stays even if all VMs have stopped.
3188  * It is only freed when the kvm-hv module is unloaded.
3189  * It's OK for this routine to fail, we just don't support host
3190  * core operations like redirecting H_IPI wakeups.
3191  */
3192 void kvmppc_alloc_host_rm_ops(void)
3193 {
3194 	struct kvmppc_host_rm_ops *ops;
3195 	unsigned long l_ops;
3196 	int cpu, core;
3197 	int size;
3198 
3199 	/* Not the first time here ? */
3200 	if (kvmppc_host_rm_ops_hv != NULL)
3201 		return;
3202 
3203 	ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3204 	if (!ops)
3205 		return;
3206 
3207 	size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3208 	ops->rm_core = kzalloc(size, GFP_KERNEL);
3209 
3210 	if (!ops->rm_core) {
3211 		kfree(ops);
3212 		return;
3213 	}
3214 
3215 	get_online_cpus();
3216 
3217 	for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3218 		if (!cpu_online(cpu))
3219 			continue;
3220 
3221 		core = cpu >> threads_shift;
3222 		ops->rm_core[core].rm_state.in_host = 1;
3223 	}
3224 
3225 	ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3226 
3227 	/*
3228 	 * Make the contents of the kvmppc_host_rm_ops structure visible
3229 	 * to other CPUs before we assign it to the global variable.
3230 	 * Do an atomic assignment (no locks used here), but if someone
3231 	 * beats us to it, just free our copy and return.
3232 	 */
3233 	smp_wmb();
3234 	l_ops = (unsigned long) ops;
3235 
3236 	if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3237 		put_online_cpus();
3238 		kfree(ops->rm_core);
3239 		kfree(ops);
3240 		return;
3241 	}
3242 
3243 	cpuhp_setup_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3244 				  "ppc/kvm_book3s:prepare",
3245 				  kvmppc_set_host_core,
3246 				  kvmppc_clear_host_core);
3247 	put_online_cpus();
3248 }
3249 
3250 void kvmppc_free_host_rm_ops(void)
3251 {
3252 	if (kvmppc_host_rm_ops_hv) {
3253 		cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3254 		kfree(kvmppc_host_rm_ops_hv->rm_core);
3255 		kfree(kvmppc_host_rm_ops_hv);
3256 		kvmppc_host_rm_ops_hv = NULL;
3257 	}
3258 }
3259 #endif
3260 
3261 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3262 {
3263 	unsigned long lpcr, lpid;
3264 	char buf[32];
3265 
3266 	/* Allocate the guest's logical partition ID */
3267 
3268 	lpid = kvmppc_alloc_lpid();
3269 	if ((long)lpid < 0)
3270 		return -ENOMEM;
3271 	kvm->arch.lpid = lpid;
3272 
3273 	kvmppc_alloc_host_rm_ops();
3274 
3275 	/*
3276 	 * Since we don't flush the TLB when tearing down a VM,
3277 	 * and this lpid might have previously been used,
3278 	 * make sure we flush on each core before running the new VM.
3279 	 * On POWER9, the tlbie in mmu_partition_table_set_entry()
3280 	 * does this flush for us.
3281 	 */
3282 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3283 		cpumask_setall(&kvm->arch.need_tlb_flush);
3284 
3285 	/* Start out with the default set of hcalls enabled */
3286 	memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3287 	       sizeof(kvm->arch.enabled_hcalls));
3288 
3289 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
3290 		kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3291 
3292 	/* Init LPCR for virtual RMA mode */
3293 	kvm->arch.host_lpid = mfspr(SPRN_LPID);
3294 	kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3295 	lpcr &= LPCR_PECE | LPCR_LPES;
3296 	lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3297 		LPCR_VPM0 | LPCR_VPM1;
3298 	kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3299 		(VRMA_VSID << SLB_VSID_SHIFT_1T);
3300 	/* On POWER8 turn on online bit to enable PURR/SPURR */
3301 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
3302 		lpcr |= LPCR_ONL;
3303 	/*
3304 	 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3305 	 * Set HVICE bit to enable hypervisor virtualization interrupts.
3306 	 */
3307 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3308 		lpcr &= ~LPCR_VPM0;
3309 		lpcr |= LPCR_HVICE;
3310 	}
3311 
3312 	kvm->arch.lpcr = lpcr;
3313 
3314 	/*
3315 	 * Work out how many sets the TLB has, for the use of
3316 	 * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3317 	 */
3318 	if (cpu_has_feature(CPU_FTR_ARCH_300))
3319 		kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;	/* 256 */
3320 	else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3321 		kvm->arch.tlb_sets = POWER8_TLB_SETS;		/* 512 */
3322 	else
3323 		kvm->arch.tlb_sets = POWER7_TLB_SETS;		/* 128 */
3324 
3325 	/*
3326 	 * Track that we now have a HV mode VM active. This blocks secondary
3327 	 * CPU threads from coming online.
3328 	 */
3329 	kvm_hv_vm_activated();
3330 
3331 	/*
3332 	 * Create a debugfs directory for the VM
3333 	 */
3334 	snprintf(buf, sizeof(buf), "vm%d", current->pid);
3335 	kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3336 	if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3337 		kvmppc_mmu_debugfs_init(kvm);
3338 
3339 	return 0;
3340 }
3341 
3342 static void kvmppc_free_vcores(struct kvm *kvm)
3343 {
3344 	long int i;
3345 
3346 	for (i = 0; i < KVM_MAX_VCORES; ++i)
3347 		kfree(kvm->arch.vcores[i]);
3348 	kvm->arch.online_vcores = 0;
3349 }
3350 
3351 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3352 {
3353 	debugfs_remove_recursive(kvm->arch.debugfs_dir);
3354 
3355 	kvm_hv_vm_deactivated();
3356 
3357 	kvmppc_free_vcores(kvm);
3358 
3359 	kvmppc_free_hpt(kvm);
3360 
3361 	kvmppc_free_pimap(kvm);
3362 }
3363 
3364 /* We don't need to emulate any privileged instructions or dcbz */
3365 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3366 				     unsigned int inst, int *advance)
3367 {
3368 	return EMULATE_FAIL;
3369 }
3370 
3371 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3372 					ulong spr_val)
3373 {
3374 	return EMULATE_FAIL;
3375 }
3376 
3377 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3378 					ulong *spr_val)
3379 {
3380 	return EMULATE_FAIL;
3381 }
3382 
3383 static int kvmppc_core_check_processor_compat_hv(void)
3384 {
3385 	if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3386 	    !cpu_has_feature(CPU_FTR_ARCH_206))
3387 		return -EIO;
3388 	/*
3389 	 * Disable KVM for Power9 in radix mode.
3390 	 */
3391 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
3392 		return -EIO;
3393 
3394 	return 0;
3395 }
3396 
3397 #ifdef CONFIG_KVM_XICS
3398 
3399 void kvmppc_free_pimap(struct kvm *kvm)
3400 {
3401 	kfree(kvm->arch.pimap);
3402 }
3403 
3404 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3405 {
3406 	return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3407 }
3408 
3409 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3410 {
3411 	struct irq_desc *desc;
3412 	struct kvmppc_irq_map *irq_map;
3413 	struct kvmppc_passthru_irqmap *pimap;
3414 	struct irq_chip *chip;
3415 	int i;
3416 
3417 	if (!kvm_irq_bypass)
3418 		return 1;
3419 
3420 	desc = irq_to_desc(host_irq);
3421 	if (!desc)
3422 		return -EIO;
3423 
3424 	mutex_lock(&kvm->lock);
3425 
3426 	pimap = kvm->arch.pimap;
3427 	if (pimap == NULL) {
3428 		/* First call, allocate structure to hold IRQ map */
3429 		pimap = kvmppc_alloc_pimap();
3430 		if (pimap == NULL) {
3431 			mutex_unlock(&kvm->lock);
3432 			return -ENOMEM;
3433 		}
3434 		kvm->arch.pimap = pimap;
3435 	}
3436 
3437 	/*
3438 	 * For now, we only support interrupts for which the EOI operation
3439 	 * is an OPAL call followed by a write to XIRR, since that's
3440 	 * what our real-mode EOI code does.
3441 	 */
3442 	chip = irq_data_get_irq_chip(&desc->irq_data);
3443 	if (!chip || !is_pnv_opal_msi(chip)) {
3444 		pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3445 			host_irq, guest_gsi);
3446 		mutex_unlock(&kvm->lock);
3447 		return -ENOENT;
3448 	}
3449 
3450 	/*
3451 	 * See if we already have an entry for this guest IRQ number.
3452 	 * If it's mapped to a hardware IRQ number, that's an error,
3453 	 * otherwise re-use this entry.
3454 	 */
3455 	for (i = 0; i < pimap->n_mapped; i++) {
3456 		if (guest_gsi == pimap->mapped[i].v_hwirq) {
3457 			if (pimap->mapped[i].r_hwirq) {
3458 				mutex_unlock(&kvm->lock);
3459 				return -EINVAL;
3460 			}
3461 			break;
3462 		}
3463 	}
3464 
3465 	if (i == KVMPPC_PIRQ_MAPPED) {
3466 		mutex_unlock(&kvm->lock);
3467 		return -EAGAIN;		/* table is full */
3468 	}
3469 
3470 	irq_map = &pimap->mapped[i];
3471 
3472 	irq_map->v_hwirq = guest_gsi;
3473 	irq_map->desc = desc;
3474 
3475 	/*
3476 	 * Order the above two stores before the next to serialize with
3477 	 * the KVM real mode handler.
3478 	 */
3479 	smp_wmb();
3480 	irq_map->r_hwirq = desc->irq_data.hwirq;
3481 
3482 	if (i == pimap->n_mapped)
3483 		pimap->n_mapped++;
3484 
3485 	kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3486 
3487 	mutex_unlock(&kvm->lock);
3488 
3489 	return 0;
3490 }
3491 
3492 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3493 {
3494 	struct irq_desc *desc;
3495 	struct kvmppc_passthru_irqmap *pimap;
3496 	int i;
3497 
3498 	if (!kvm_irq_bypass)
3499 		return 0;
3500 
3501 	desc = irq_to_desc(host_irq);
3502 	if (!desc)
3503 		return -EIO;
3504 
3505 	mutex_lock(&kvm->lock);
3506 
3507 	if (kvm->arch.pimap == NULL) {
3508 		mutex_unlock(&kvm->lock);
3509 		return 0;
3510 	}
3511 	pimap = kvm->arch.pimap;
3512 
3513 	for (i = 0; i < pimap->n_mapped; i++) {
3514 		if (guest_gsi == pimap->mapped[i].v_hwirq)
3515 			break;
3516 	}
3517 
3518 	if (i == pimap->n_mapped) {
3519 		mutex_unlock(&kvm->lock);
3520 		return -ENODEV;
3521 	}
3522 
3523 	kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3524 
3525 	/* invalidate the entry */
3526 	pimap->mapped[i].r_hwirq = 0;
3527 
3528 	/*
3529 	 * We don't free this structure even when the count goes to
3530 	 * zero. The structure is freed when we destroy the VM.
3531 	 */
3532 
3533 	mutex_unlock(&kvm->lock);
3534 	return 0;
3535 }
3536 
3537 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3538 					     struct irq_bypass_producer *prod)
3539 {
3540 	int ret = 0;
3541 	struct kvm_kernel_irqfd *irqfd =
3542 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3543 
3544 	irqfd->producer = prod;
3545 
3546 	ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3547 	if (ret)
3548 		pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3549 			prod->irq, irqfd->gsi, ret);
3550 
3551 	return ret;
3552 }
3553 
3554 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3555 					      struct irq_bypass_producer *prod)
3556 {
3557 	int ret;
3558 	struct kvm_kernel_irqfd *irqfd =
3559 		container_of(cons, struct kvm_kernel_irqfd, consumer);
3560 
3561 	irqfd->producer = NULL;
3562 
3563 	/*
3564 	 * When producer of consumer is unregistered, we change back to
3565 	 * default external interrupt handling mode - KVM real mode
3566 	 * will switch back to host.
3567 	 */
3568 	ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3569 	if (ret)
3570 		pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3571 			prod->irq, irqfd->gsi, ret);
3572 }
3573 #endif
3574 
3575 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3576 				 unsigned int ioctl, unsigned long arg)
3577 {
3578 	struct kvm *kvm __maybe_unused = filp->private_data;
3579 	void __user *argp = (void __user *)arg;
3580 	long r;
3581 
3582 	switch (ioctl) {
3583 
3584 	case KVM_PPC_ALLOCATE_HTAB: {
3585 		u32 htab_order;
3586 
3587 		r = -EFAULT;
3588 		if (get_user(htab_order, (u32 __user *)argp))
3589 			break;
3590 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
3591 		if (r)
3592 			break;
3593 		r = -EFAULT;
3594 		if (put_user(htab_order, (u32 __user *)argp))
3595 			break;
3596 		r = 0;
3597 		break;
3598 	}
3599 
3600 	case KVM_PPC_GET_HTAB_FD: {
3601 		struct kvm_get_htab_fd ghf;
3602 
3603 		r = -EFAULT;
3604 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
3605 			break;
3606 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3607 		break;
3608 	}
3609 
3610 	default:
3611 		r = -ENOTTY;
3612 	}
3613 
3614 	return r;
3615 }
3616 
3617 /*
3618  * List of hcall numbers to enable by default.
3619  * For compatibility with old userspace, we enable by default
3620  * all hcalls that were implemented before the hcall-enabling
3621  * facility was added.  Note this list should not include H_RTAS.
3622  */
3623 static unsigned int default_hcall_list[] = {
3624 	H_REMOVE,
3625 	H_ENTER,
3626 	H_READ,
3627 	H_PROTECT,
3628 	H_BULK_REMOVE,
3629 	H_GET_TCE,
3630 	H_PUT_TCE,
3631 	H_SET_DABR,
3632 	H_SET_XDABR,
3633 	H_CEDE,
3634 	H_PROD,
3635 	H_CONFER,
3636 	H_REGISTER_VPA,
3637 #ifdef CONFIG_KVM_XICS
3638 	H_EOI,
3639 	H_CPPR,
3640 	H_IPI,
3641 	H_IPOLL,
3642 	H_XIRR,
3643 	H_XIRR_X,
3644 #endif
3645 	0
3646 };
3647 
3648 static void init_default_hcalls(void)
3649 {
3650 	int i;
3651 	unsigned int hcall;
3652 
3653 	for (i = 0; default_hcall_list[i]; ++i) {
3654 		hcall = default_hcall_list[i];
3655 		WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3656 		__set_bit(hcall / 4, default_enabled_hcalls);
3657 	}
3658 }
3659 
3660 static struct kvmppc_ops kvm_ops_hv = {
3661 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3662 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3663 	.get_one_reg = kvmppc_get_one_reg_hv,
3664 	.set_one_reg = kvmppc_set_one_reg_hv,
3665 	.vcpu_load   = kvmppc_core_vcpu_load_hv,
3666 	.vcpu_put    = kvmppc_core_vcpu_put_hv,
3667 	.set_msr     = kvmppc_set_msr_hv,
3668 	.vcpu_run    = kvmppc_vcpu_run_hv,
3669 	.vcpu_create = kvmppc_core_vcpu_create_hv,
3670 	.vcpu_free   = kvmppc_core_vcpu_free_hv,
3671 	.check_requests = kvmppc_core_check_requests_hv,
3672 	.get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3673 	.flush_memslot  = kvmppc_core_flush_memslot_hv,
3674 	.prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3675 	.commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3676 	.unmap_hva = kvm_unmap_hva_hv,
3677 	.unmap_hva_range = kvm_unmap_hva_range_hv,
3678 	.age_hva  = kvm_age_hva_hv,
3679 	.test_age_hva = kvm_test_age_hva_hv,
3680 	.set_spte_hva = kvm_set_spte_hva_hv,
3681 	.mmu_destroy  = kvmppc_mmu_destroy_hv,
3682 	.free_memslot = kvmppc_core_free_memslot_hv,
3683 	.create_memslot = kvmppc_core_create_memslot_hv,
3684 	.init_vm =  kvmppc_core_init_vm_hv,
3685 	.destroy_vm = kvmppc_core_destroy_vm_hv,
3686 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3687 	.emulate_op = kvmppc_core_emulate_op_hv,
3688 	.emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3689 	.emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3690 	.fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3691 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3692 	.hcall_implemented = kvmppc_hcall_impl_hv,
3693 #ifdef CONFIG_KVM_XICS
3694 	.irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3695 	.irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3696 #endif
3697 };
3698 
3699 static int kvm_init_subcore_bitmap(void)
3700 {
3701 	int i, j;
3702 	int nr_cores = cpu_nr_cores();
3703 	struct sibling_subcore_state *sibling_subcore_state;
3704 
3705 	for (i = 0; i < nr_cores; i++) {
3706 		int first_cpu = i * threads_per_core;
3707 		int node = cpu_to_node(first_cpu);
3708 
3709 		/* Ignore if it is already allocated. */
3710 		if (paca[first_cpu].sibling_subcore_state)
3711 			continue;
3712 
3713 		sibling_subcore_state =
3714 			kmalloc_node(sizeof(struct sibling_subcore_state),
3715 							GFP_KERNEL, node);
3716 		if (!sibling_subcore_state)
3717 			return -ENOMEM;
3718 
3719 		memset(sibling_subcore_state, 0,
3720 				sizeof(struct sibling_subcore_state));
3721 
3722 		for (j = 0; j < threads_per_core; j++) {
3723 			int cpu = first_cpu + j;
3724 
3725 			paca[cpu].sibling_subcore_state = sibling_subcore_state;
3726 		}
3727 	}
3728 	return 0;
3729 }
3730 
3731 static int kvmppc_book3s_init_hv(void)
3732 {
3733 	int r;
3734 	/*
3735 	 * FIXME!! Do we need to check on all cpus ?
3736 	 */
3737 	r = kvmppc_core_check_processor_compat_hv();
3738 	if (r < 0)
3739 		return -ENODEV;
3740 
3741 	r = kvm_init_subcore_bitmap();
3742 	if (r)
3743 		return r;
3744 
3745 	/*
3746 	 * We need a way of accessing the XICS interrupt controller,
3747 	 * either directly, via paca[cpu].kvm_hstate.xics_phys, or
3748 	 * indirectly, via OPAL.
3749 	 */
3750 #ifdef CONFIG_SMP
3751 	if (!get_paca()->kvm_hstate.xics_phys) {
3752 		struct device_node *np;
3753 
3754 		np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
3755 		if (!np) {
3756 			pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
3757 			return -ENODEV;
3758 		}
3759 	}
3760 #endif
3761 
3762 	kvm_ops_hv.owner = THIS_MODULE;
3763 	kvmppc_hv_ops = &kvm_ops_hv;
3764 
3765 	init_default_hcalls();
3766 
3767 	init_vcore_lists();
3768 
3769 	r = kvmppc_mmu_hv_init();
3770 	return r;
3771 }
3772 
3773 static void kvmppc_book3s_exit_hv(void)
3774 {
3775 	kvmppc_free_host_rm_ops();
3776 	kvmppc_hv_ops = NULL;
3777 }
3778 
3779 module_init(kvmppc_book3s_init_hv);
3780 module_exit(kvmppc_book3s_exit_hv);
3781 MODULE_LICENSE("GPL");
3782 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3783 MODULE_ALIAS("devname:kvm");
3784 
3785